Infrastructure Messages

Services in the Dragon runtime interact with each other using messages transported with a variety of different means (mostly Channels). Although typically there will be a Python API to construct and send these messages, the messages themselves constitute the true internal interface. To that end, they are a convention.

Here we document this internal message API.

Generalities about Messages

Many messages are part of a request/response pair. At the level discussed here, successful and error responses are carried by the same response type but the fields present or valid may change according to whether the response is successful or an error.

All messages are serialized using JSON; the reason for this is to allow non-Python actors to interact with the Dragon infrastructure as well as allowing the possibility for the different components of the Dragon runtime to be implemented in something other than Python.

The serialization method may change in the future as an optimization; JSON is chosen here as a reasonable cross language starting point. In particular, the fields discussed here for basic message parsing (type and instance tagging, and errors) are defined in terms of integer valued enumerations suitable for a fixed field location and width binary determination.

The top level of the JSON structure will be a dictionary (map) where the ‘_tc’ key’s value is an integer corresponding to an element of the dragon.messages.MsgTypes enumeration class. The other fields of this map will be defined on a message by message basis.

The canonical form of the message will be taken on the JSON level, not the string level. This means that messages should not be compared or sorted as strings. Internally the Python interface will construct inbound messages into class objects of distinct type according to the ‘_tc’ field in the initializer by using a factory function. The reason to have a lot of different message types instead of just one type differentiated by the value of the ‘_tc’ field is that this allows IDEs and documentation tools to work better by having explicit knowledge of what fields and methods are expected in any particular message.

Common Fields

These fields are common to every message described below depending on their category of classification.

The _tc TypeCode Field

The _tc typecode field is used during parsing to determine the type of a received message. The message format is JSON. The _tc field can be used to determine of all expected fields of a message are indeed in the message to verify its format.

Beyond the _tc typecode field, there are other fields expected to belong to every message.

The tag Field

Every message has a 64 bit positive unsigned integer tag field. Together with the identity of the sender is implicit or explicitly identified in some other field, this serves to uniquely identify the message.

Ideally, this would be throughout the life of the runtime, but it’s enough to say that no two messages should be simultaneously relevant with the same (sender, tag).

Message Categories

There are three general categories of messages:

request messages, where one entity asks another one to do something
response messages, returning the result to the requester
- the name of these messages normally ends “Response”
other messages, mostly to do with infrastructure sequencing

Every request message will contain p_uid and r_c_uid fields. These fields denote the unique process ID of the entity that created the request and the unique channel ID to send the response to. See Conventional IDs for more details on process identification.

The ref Field

Every response message generated in reply to a request message will contain a ref field that echos the tag field of the request message that caused it to be generated. If a response message is generated asynchronously - for instance, as part of a startup notification or some event such as a process exiting unexpectedly - then the ref field will be 0, which is an illegal value for the tag field.

The err Field

Every response message will also have an err field, holding an integer enumeration particular to the response. This enumeration will be the class attribute Errors in the response message class. The enumeration element with value 0 will always indicate success. Which fields are present in the response message may depend on the enumeration value.

Format of Messages List

Class Name

type enum
member of dragon.messages.MsgTypes enum class

purpose
succinct description of purpose, sender and receiver, where and when typically seen. Starts with ‘Request’ if it is a request message and ‘Response’ if it is a response message.

fields
Fields in the message, together with description of how they are to get parsed out. Some fields may themselves be dictionary structures used to initialize member objects.

Fields in common to every message (such as tag, ref, and err) are not mentioned here.

An expected type will be mentioned on each field where it is an obvious primitive (such as a string or an integer). Where a type isn’t mentioned, it is assumed to be another key-value map to be interpreted as the initializer for some more complex type.

response
If a request kind of message, gives the class name of the corresponding response message, if applicable

request
If a request kind of message, gives the class name of the corresponding response message.

see also
Other related messages

Global Services Messages API

These messages are the ones underlying the Global Services API and are sent only to and from Global Services.

All global services request messages have an integer p_uid field that contains the process UID of the requester and an integer r_c_uid field that contains the channel ID to send the response to. This field is not echoed in the corresponding response. For details on IDs, see Conventional IDs.

GS Process Messages

GSProcessCreate
type enum
GS_PROCESS_CREATE

purpose
Request to global services to create a new managed process.

fields
exe

string

executable name

args

list of strings

arguments to the executable. [exe, args] == argv

env

map, key=string value=string

Environment variables to be exported before running.

Note that this is additive/override to the shepherd environment - may need to be more fancy here if it turns out we need to add to paths

rundir

string, absolute path

Indicates where the process is to be run from. If empty, does not matter.

user_name

string

requested user name for the process

options

options

initializer for other process options object

output_c_uid is the channel id where output, both stdout and stderr, should be

forwarded if present in an SHFwdOutput message. If this field is not present in the options, then output will be forwarded to the Backend. The SHFwdOutput message indicates whether it was from stdout or stderr.

response GSProcessCreateResponse
see also
GSProcessDestroy

refer to the Common Fields section for additional request message fields

implementation(s): Python

GSProcessCreateResponse
type enum
GS_PROCESS_CREATE_RESPONSE

purpose
Response to process creation request.

fields
Alternatives on err:

SUCCESS ( = 0)
A new process was created

desc

map

initializer for ProcessDescriptor - includes at a minimum p_uid of new process and assigned user name.

FAIL
No process was created

err_info

string

explanation of what went wrong

Might add some more here, e.g. is this a GS related problem or a Shepherd related problem.

ALREADY
The user name was already in use

desc

map

init for the ProcessDescriptor of the process that already exists with that user name

request
GSProcessCreate
implementation(s): Python

GSProcessList

type enum
GS_PROCESS_LIST

purpose
Return a list of the p_uid for all the processes currently being managed

fields
None additional

response
GSProcessListResponse

see also
GSProcessQuery

refer to the Common Fields section for additional request message fields

implementation(s): Python

GSProcessListResponse
type enum
GS_PROCESS_LIST_RESPONSE

purpose
Responds with a list of the p_uid for all the processes currently being managed

fields
plist

list of nonnegative integers for all the processes
request
GSProcessList

see also
GSProcessQuery
implementation(s): Python

GSProcessQuery
type enum
GS_PROCESS_QUERY

purpose
Request the ProcessDescriptor for a managed process

fields
One of the t_p_uid and user_name fields must be present.

If both are specified, then the user_name field is ignored.

t_p_uid

integer

target process UID

user_name

string

user supplied name for the target process
response
GSProcessQueryResponse

refer to the Common Fields section for additional request message fields
implementation(s): Python

GSProcessQueryResponse
type enum
GS_PROCESS_QUERY_RESPONSE

purpose
Response to request for ProcessDescriptor for a managed process

fields
Alternatives on err:

SUCCESS (= 0)
Process has been found

desc

map

initializer for ProcessDescriptor - includes at a minimum p_uid of new process and assigned user name.

UNKNOWN (= 1)
No such process is known.

err_info

string

explanation of what went wrong
request
GSProcessQuery

see also
GSProcessList
implementation(s): Python

GSProcessKill
type enum
GS_PROCESS_KILL

purpose
Request a managed process get killed

fields
One of the t_p_uid and user_name fields must be present.

If both are specified, then the user_name field is ignored.

t_p_uid

integer

target process UID

user_name

string

user supplied name for the target process

sig

integer

signal to kill the process with
response
GSProcessKillResponse

see also
refer to the Common Fields section for additional request message fields
implementation(s): Python

GSProcessKillResponse
type enum
GS_PROCESS_KILL_RESPONSE

purpose
Response to GSProcessKill message

fields
Alternatives on err:

SUCCESS (= 0)
Process has been found and has been killed.

exit_code

integer

exit code the process returned

UNKNOWN (= 1)
No such process is known.

err_info

string

explanation of what went wrong

FAIL_KILL (= 2)
Something went wrong in killing the process.

err_info

string

explanation of what went wrong

DEAD (= 3)
The process is already dead.

No other fields.
request
GSProcessKill

see also
GSProcessQuery
implementation(s): Python

GSProcessJoin
type enum
GS_PROCESS_JOIN

purpose
Request notification when a given process exits

fields
One of the t_p_uid and user_name fields must be present.

If both are specified, then the user_name field is ignored.

t_p_uid

integer

target process UID

user_name

string

user supplied name for the target process

timeout

integer, interpreted as microseconds

timeout value; any value < 0 interpreted as infinite timeout
response
GSProcessJoinResponse

see also
refer to the Common Fields section for additional request message fields
implementation(s): Python

GSProcessJoinResponse
type enum
GS_PROCESS_JOIN_RESPONSE

purpose
Response to request for notification when a process exits.

fields
Alternatives on err:
SUCCESS (= 0)
Target process has exited. If the target process has already exited, this is not an error.

exit_code

integer

the Unix return value of the process
UNKNOWN (= 1)
No such process is known.

err_info

string

explanation of what went wrong
TIMEOUT (= 2)
The timer has expired and the process still hasn’t exited.

No fields.
request
GSProcessJoin

see also
GSProcessQuery
implementation(s): Python

GSProcessJoinList
type enum
GS_PROCESS_JOIN_LIST

purpose
Request notification when any/all process from a given list exits.

fields
Both t_p_uid_list and user_name_list fields could be present. Any process in the list could be identified by either p_uid or user_name.
t_p_uid_list
list of integers

list of target processes UID
user_name_list
list of strings

list of user supplied names for the target processes
timeout
integer, interpreted as microseconds

timeout value; any value < 0 interpreted as infinite timeout
join_all
bool

default False, request notification when any process from the list exits. When True, request notification when all processes exit.
response
GSProcessJoinListResponse

see also
refer to the Common Fields section for additional request message fields
implementation(s): Python

GSProcessJoinListResponse
type enum
GS_PROCESS_JOIN_LIST_RESPONSE

purpose
Response to request for notification when any/all process from a given list exits.

fields
Reflect the status of every process in the list:
**puid_status*
dictionary with the status of every process

key: p_uid

value: tuple (status, info)
The status of a process in the list can be one of the following:
SUCCESS (= 0)
Target process has exited. If the target process has already exited, this is not an error.

info:

integer

the Unix return value of the process
UNKNOWN (= 1)
No such process is known.

info:

string

explanation of what went wrong
TIMEOUT (= 2)
The timer has expired and the process still hasn’t exited.

info:
None

SELF (= 3)
Attempt to join itself.

info:

string

explanation of what went wrong
PENDING (= 4)
The process is still pending (not exited, no timeout).

info:
None
request
GSProcessJoinList
implementation(s): Python

Pool Messages

These messages concern the lifecycle of named memory pools. All requests related to channels will include the p_uid field of the requesting entity and the r_c_uid field for the channel to send the response to.

Pools are identified in the runtime system with a nonnegative globally unique ID number, the ‘memory UID’ normally abbreviated as m_uid.

GSPoolCreate

type enum
GS_POOL_CREATE

purpose
Requests that a new user memory pool be created.

fields
user_name
string

optional user specified name for the pool. If absent or empty, Global Services will select a name.

this may not necessarily be the same name used on the node to name the pool object to the OS
options
map

Initializer for globalservices.pool.PoolOptions object. This is an open type intended to encompass what is in a dragonMemoryPoolAttr_t but also will carry system level options such as what node to create the pool on.
size
integer > 0

size of pool requested, in bytes. Note that the actual size of the pool delivered may be larger than this
response
GSPoolCreateResponse

see also
SHPoolCreate

refer to the Common Fields section for additional request message fields

implementation(s): Python

GSPoolCreateResponse
type enum
GS_POOL_CREATE_RESPONSE

purpose
Response to request for a pool creation

fields
Alternatives on err:
SUCCESS (= 0)
desc

map

Initializer for a globalservices.pool.PoolDescriptor object. This object will contain the assigned m_uid and name for the pool, as well as the (serialized) dragonMemoryPoolSerial_t library descriptor.
FAIL (= 1)
Something went wrong in pool creation. This can be an error on the node level (found when the call to allocate the resources is actually made) or a logical error with the request.

err_info

string

explanation of what went wrong

err_code

integer

The error code from the library call, if applicable. Not all errors will be revealed when the call is made; if the error isn’t due to the underlying API call
ALREADY (= 2)
The user name for the pool was already in use; the current descriptor is returned.

desc

map

Initializer for a globalservices.pool.PoolDescriptor object. This object will contain the assigned m_uid and name for the pool as well as the (serialized) dragonMemoryPoolSerial_t library descriptor.
request
GSPoolCreate

see also
SHPoolCreate, SHPoolCreateResponse
implementation(s): Python

GSPoolDestroy
type enum
GS_POOL_DESTROY

purpose
Request destruction of a managed memory pool.

fields
One of the m_uid and user_name fields must be present.

If both are specified, then the user_name field is ignored.
m_uid
integer

target memory pool ID
user_name
string

user supplied name for the target pool
response
GSPoolDestroyResponse

see also
refer to the Common Fields section for additional request message fields
implementation(s): Python

GSPoolDestroyResponse
type enum
GS_POOL_DESTROY_RESPONSE

purpose
Response to GSPoolDestroy message

fields
Alternatives on err:
SUCCESS (= 0)
Pool has been found and has been removed.

No other fields.

UNKNOWN (= 1)
No such pool is known.

err_info

string

explanation of what went wrong
FAIL (= 2)
Something went wrong in removing the pool. This could be for a number of reasons.

err_info

string

explanation of what went wrong

err_code

integer

The error code from the library call, if applicable. Not all errors will be revealed when the call is made; if the error isn’t due to the underlying API call
GONE (= 3)
The pool is already destroyed

No other fields.

PENDING (= 4)
Pool creation is pending, and it can’t be destroyed. Wait until the GSPoolCreateResponse message has been received.
request
GSPoolDestroy
implementation(s): Python

GSPoolList

type enum
GS_POOL_LIST

purpose
Return a list of tuples of m_uid for all pools currently alive.

fields
None additional

response
GSPoolListResponse

see also
GSPoolQuery

refer to the Common Fields section for additional request message fields

implementation(s): Python

GSPoolListResponse
type enum
GS_POOL_LIST_RESPONSE

purpose
Responds with a list of m_uid for all the pools currently alive

fields
mlist
list of nonnegative integers
request
GSPoolList

see also
GSPoolQuery
implementation(s): Python

GSPoolQuery
type enum
GS_POOL_QUERY

purpose
Request the PoolDescriptor for a managed memory pool.

fields
One of the m_uid and user_name fields must be present.

If both are specified, then the user_name field is ignored.
m_uid
integer

target pool UID
user_name
string

user supplied name for the target pool
response
GSPoolQueryResponse

refer to the Common Fields section for additional request message fields
implementation(s): Python

GSPoolQueryResponse
type enum
GS_POOL_QUERY_RESPONSE

purpose
Response to request for PoolDescriptor for a managed memory pool. This object carries a serialized library pool descriptor and which node the pool is found on.

fields
Alternatives on err:
SUCCESS (= 0)
Pool has been found

desc

map

initializer for PoolDescriptor
UNKNOWN (= 1)
No such pool is known.

err_info

string

explanation of what went wrong
request
GSPoolQuery

see also
GSPoolList
implementation(s): Python

Channel Messages

These messages are related to the lifecycle of channels. All requests related to channels will include a p_uid field, of the requesting process and the r_c_uid field for the channel to send the response to.

Channels are identified in the runtime system with a nonnegative globally unique ID number, the ‘channel UID’ normally abbreviated as c_uid. Note that every channel is associated with a memory pool which itself has its own m_uid.

GSChannelCreate
type enum
GS_CHANNEL_CREATE

purpose
Requests that a new channel be created.

fields
user_name
string

optional user specified name for the channel. If absent or empty, Global Services will select a name.
options
map

Initializer for a (global) ChannelOptions object. This is an open type intended to completely describe the functionality of the channel. It contains fields pertaining to the global function and local characteristics of the channel.
m_uid
integer

m_uid of the pool to create the channel in

if 0, use the default pool on the target node.
response
GSChannelCreateResponse

see also
GSChannelQuery

refer to the Common Fields section for additional request message fields

implementation(s): Python
Python

GSChannelCreateResponse
type enum
GS_CHANNEL_CREATE_RESPONSE

purpose
Response to channel creation request.

fields
Alternatives on err:
SUCCESS (= 0)
desc

map

Initializer for a ChannelDescriptor object. This object will contain the assigned c_uid and name for the channel and the serialized channel descriptor as well as the m_uid for the pool associated with the channel.
FAIL (= 1)
Something went wrong in channel creation.

err_info

string

explanation of what went wrong
ALREADY (= 2)
The user name was already in use

desc

map

Initializer for a ChannelDescriptor object. This object will contain the assigned c_uid and name for the channel as well as the serialized channel descriptor.
request
GSChannelCreate
implementation(s): Python

GSChannelList

type enum
GS_CHANNEL_LIST

purpose
Request list of currently active channels.

fields
None other than mandatory fields. To-do: add something to make this more selective.

response
GSChannelListResponse

see also
GSChannelQuery

refer to the Common Fields section for additional request message fields

implementation(s): Python

GSChannelListResponse
type enum
GS_CHANNEL_LIST_RESPONSE

purpose
Response to request to list of currently active channels with a list of tuples of (c_uid, name) for all the channels currently being managed

fields
clist
list of integers

The list contains the currently active c_uids
request
GSChannelList

see also
GSChannelQuery
implementation(s): Python

GSChannelQuery
type enum
GS_CHANNEL_QUERY

purpose
Request the descriptor for an already created channel by channel UID or user name.

fields
One of the c_uid and user_name fields must be present.

If both are specified, then the user_name field is ignored.
c_uid
integer

target channel UID
user_name
string

user supplied name for the target channel
inc_refcnt
bool

whether to increment refcnt on the Channel. Default False.
response
GSChannelQueryResponse

see also
refer to the Common Fields section for additional request message fields
implementation(s): Python

GSChannelQueryResponse
type enum
GS_CHANNEL_QUERY_RESPONSE

purpose
Response to request for a channel descriptor

fields
Alternatives on err:
SUCCESS (= 0)
desc

map

Initializer for a ChannelDescriptor object. This object will contain the assigned c_uid and name for the channel as well as the m_uid of the pool associated with the channel. It also includes the serialized library level descriptor.
UNKNOWN (= 1)
No such channel is active.
request
GSChannelQuery
implementation(s): Python

GSChannelDestroy
type enum
GS_CHANNEL_DESTROY

purpose
Request that a channel be destroyed or a refcount on the channel be released.

fields
One of the c_uid and user_name fields must be present.

If both are specified, then the user_name field is ignored.
c_uid
integer

target channel UID
user_name
string

user supplied name for the target channel
reply_req
bool

default True, reply to this message requested
dec_ref
bool

default False, release refcount on this channel
response
GSChannelDestroyResponse

see also
refer to the Common Fields section for additional request message fields
implementation(s): Python

GSChannelDestroyResponse
type enum
GS_CHANNEL_DESTROY_RESPONSE

purpose
Response to request to destroy a channel.

fields
Alternatives on err:
SUCCESS (= 0)
No fields.

UNKNOWN (= 1)
No such channel is known.

FAIL (= 2)
The channel exists but

err_info

string

explanation of what went wrong

err_code

integer

error code from the library call that failed. If the problem is not related to a library call, this value is 0.
BUSY (= 3)
Some entity is currently using the channel in some way that prevents its destruction.

err_info

string

explanation of what went wrong
request
GSChannelDestroy
implementation(s): Python

GSChannelJoin

Request to be notified when a channel with a given name is created.

Note: when we refactor the messages, we may want to get rid of GSChannelJoin and its response and ride all this on top of GSChannelQuery by adding a timeout.
type enum
GS_CHANNEL_JOIN

purpose
Sometimes two processes want to communicate through a channel but we want to create the channel lazily with either the send side or receive side doing it. This lets the other side meet the creation by name in global services without polling.

fields
identifier
string

name of channel to join to
timeout
integer, interpreted as microseconds

timeout value; any value < 0 interpreted as infinite timeout
response
GSChannelJoinResponse

refer to the Common Fields section for additional request message fields
implementation(s): Python

GSChannelJoinResponse
type enum
GS_CHANNEL_JOIN_RESPONSE

purpose
Response to request to join a channel

fields
Alternatives on err:
SUCCESS (= 0)
desc

map

Initializer for a ChannelDescriptor object. This object will contain the assigned c_uid and name for the channel as well as the m_uid of the pool associated with the channel. It also includes the serialized library level descriptor.
TIMEOUT (= 1)
The specified timeout has expired.

No fields

DEAD (= 2)
The channel used to exist, but it has been destroyed.
request
GSChannelJoin
implementation(s): Python

GSChannelDetach

PLACEHOLDER

This probably needs to mean giving back all the send and receive handles currently acquired.
type enum
GS_CHANNEL_DETACH

purpose
Request local detachment from channel.

fields
c_uid
integer

id of channel to detach from
response
GSChannelDetachResponse

see also
GSChannelAttach

refer to the Common Fields section for additional request message fields
implementation(s): Python

GSChannelDetachResponse

PLACEHOLDER

type enum
GS_CHANNEL_DETACH_RESPONSE

purpose
Response to request to detach from a channel.

fields
Alternatives on err:

SUCCESS (= 0)
No fields.

UNKNOWN (= 1)
No such process is known.

No fields

UNKNOWN_CHANNEL (= 2)
No such channel is known.

No fields

NOT_ATTACHED (= 3)
Requested to detach from something you weren’t attached to to begin with.

request
GSChannelDetach

implementation(s): Python

GSChannelGetSendh

PLACEHOLDER
type enum
GS_CHANNEL_GET_SENDH

purpose
Request send handle for channel

fields
c_uid
integer

id of channel to get a send handle for
response
GSChannelGetSendhResponse

see also
GSChannelGetRecvh

see also
refer to the Common Fields section for additional request message fields
implementation(s): Python

GSChannelGetSendhResponse

PLACEHOLDER
type enum
GS_CHANNEL_GET_SENDH_RESPONSE

purpose
Response to request to get a channel send handle

fields
Alternatives on err:
SUCCESS (= 0)
sendh

map

info needed to set up the send handle locally
UNKNOWN (= 1)
No such process is known.

No fields

UNKNOWN_CHANNEL (= 2)
No such channel is known.

No fields

NOT_ATTACHED (= 3)
Descriptor not attached, can’t get send handle.

CANT (= 4)
Can’t get a recv handle - too many already gotten or some other reason related to the type.

err_info

string

explanation of why not
request
GSChannelGetSendh
implementation(s): Python

GSChannelGetRecvh

PLACEHOLDER
type enum
GS_CHANNEL_GET_RECVH

purpose
Request recv handle for channel

fields
c_uid
integer

id of channel to get a recv handle for
response
GSChannelGetRecvhResponse

see also
GSChannelGetSendh

see also
refer to the Common Fields section for additional request message fields
implementation(s): Python

GSChannelGetRecvhResponse

PLACEHOLDER
type enum
GS_CHANNEL_GET_RECVH_RESPONSE

purpose
Response to request to get a channel recv handle

fields
Alternatives on err:
SUCCESS (= 0)
recvh

map

info needed to set up the recv handle locally
UNKNOWN (= 1)
No such process is known.

No fields

UNKNOWN_CHANNEL (= 2)
No such channel is known.

No fields

NOT_ATTACHED (= 3)
Descriptor not attached, can’t get recv handle.

CANT (= 4)
Can’t get a recv handle - too many already gotten or some other reason related to the type.

err_info

string

explanation of why not
request
GSChannelGetRecvh
implementation(s): Python

Node Messages

These messages are related to hardware transactions with Global Services, e.g. querying the number of cpus in the system.

GSNodeList

type enum
GS_NODE_LIST

purpose
Return a list of tuples of h_uid for all nodes currently registered.

fields
None additional

response
GSNodeListResponse

see also
GSNodeQuery

see also
refer to the Common Fields section for additional request message fields

implementation(s): Python

GSNodeListResponse

type enum
GS_NODE_LIST_RESPONSE

purpose
Responds with a list of h_uid for all the nodes currently registered.

fields

hlist

list of nonnegative integers

request
GSNodeList

see also
GSNodeQuery

implementation(s): Python

GSNodeQuery

type enum
GS_NODE_QUERY

purpose
Ask Global Services for a node descriptor of the hardware Dragon is running on.

fields
name
str

Name of the node to query
h_uid
int

Unique host identifier (h_uid) held by GS.

The host running the Launcher Frontend has h_uid 0

The host running Global Services has h_uid 1
response
GSNodeQueryResponse

implementation(s): Python

GSNodeQueryResponse
type enum
GS_NODE_QUERY_RESPONSE

purpose
Return the machine descriptor after a GSNodeQuery.

fields
Alternatives on err:

SUCCESS (= 0)
The machine descriptor was successfully constructed

desc

NodeDescriptor

Contains the status of the node Dragon is running on.
UNKNOWN ( = 1)
An unknown error has occured.
see also
GSMachineQuery
implementation(s): Python

GSNodeQueryAll

type enum
GS_NODE_QUERY_ALL

purpose
Asks Global Services to return the NodeDescriptor for every node that is part of the system.

fields

response
GSNodeQueryAllResponse

implementation(s): Python

GSNodeQueryAllResponse
type enum
GS_NODE_QUERY_ALL_RESPONSE

purpose
Return the machine descriptor for every node after GSNodeQueryAll.

fields
Alternatives on err:

SUCCESS (= 0)
The machine descriptor was successfully constructed

descriptors

List of NodeDescriptors, one for each node in the system
UNKNOWN ( = 1)
An unknown error has occured.
see also
GSMachineQuery
implementation(s): Python

GSNodeQueryTotalCPUCount

type enum
GS_NODE_QUERY_TOTAL_CPU_COUNT

purpose
Asks GS to return the total number of CPUS beloging to all of the registered nodes.

response
GSNodeQueryTotalCPUCountResponse

see also
refer to the Common Fields section for additional request message fields

implementation(s): Python

GSNodeQueryTotalCPUCountResponse

type enum
GS_NODE_QUERY_TOTAL_CPU_COUNT_RESPONSE

purpose
Return the total number of CPUS beloging to all of the registered nodes.

fields
Alternatives on err:

SUCCESS (= 0)

The machine descriptor was successfully constructed

total_cpus

total number of CPUS beloging to all of the registered nodes.

UNKNOWN ( = 1)
An unknown error has occured.

request
GSNodeQueryTotalCPUCount

see also
GSNodeQuery

implementation(s): Python

GSGroupDestroy

type enum
GS_GROUP_DESTROY

purpose
Ask Global Services to destroy a group of resources. This means destroy all the member-resources, as well as the container/group.

fields

g_uid

integer

group UID

user_name

string

user supplied name for the group

response
GSGroupDestroyResponse

implementation(s): Python

GSGroupDestroyResponse

type enum
GS_GROUP_DESTROY_RESPONSE

purpose
Response to GSGroupDestroy message

fields
Alternatives on err:

SUCCESS (= 0)
Group has been found and has been destroyed.

desc

base64 encoded string

The group descriptor

UNKNOWN (= 1)
No such group is known.

err_info

string

explanation of what went wrong

DEAD (= 2)
The group is already dead.

No other fields.

PENDING (= 3)
The group is pending.

err_info

string

informing message that the group is still pending

request
GSGroupDestroy

implementation(s): Python

GSGroupAddTo

type enum
GS_GROUP_DESTROY

purpose
Ask Global Services to add specific resources to an existing group of resources. The resources are already created.

fields

g_uid

integer

group UID

user_name

string

user supplied name for the group

items

list of strings or integers

list of user supplied names or UIDs for the resources

response
GSGroupAddToResponse

implementation(s): Python

GSGroupAddToResponse

type enum
GS_GROUP_ADD_TO_RESPONSE

purpose
Response to GSGroupAddTo message

fields
Alternatives on err:

SUCCESS (= 0)
Group has been found and the resources have been added to it.

desc

base64 encoded string

The group descriptor

UNKNOWN (= 1)
No such group is known.

err_info

string

explanation of what went wrong

FAIL (= 2)
The request has failed. This can happen when at least one of the resources to be added to the group is dead or not found.

err_info

string

informing message with the UIDs of the resources that were not found or are dead

DEAD (= 3)
The group is already dead.

No other fields.

PENDING (= 4)
The group is pending.

err_info

string

informing message that the group is still pending

request
GSGroupAddTo

implementation(s): Python

GSGroupRemoveFrom

type enum
GS_GROUP_REMOVE_FROM

purpose
Ask Global Services to remove specific resources from an existing group of resources.

fields

g_uid

integer

group UID

user_name

string

user supplied name for the group

items

list of strings or integers

list of user supplied names or UIDs for the resources

response
GSGroupRemoveFromResponse

implementation(s): Python

GSGroupRemoveFromResponse

type enum
GS_GROUP_REMOVE_FROM_RESPONSE

purpose
Response to GSGroupRemoveFrom message

fields
Alternatives on err:

SUCCESS (= 0)
Group has been found and the resources have been removed from it.

desc

base64 encoded string

The group descriptor

UNKNOWN (= 1)
No such group is known.

err_info

string

explanation of what went wrong

FAIL (= 2)
The request has failed. This can happen when at least one of the resources to be removed from the group was not found.

err_info

string

informing message with the UIDs of those resources that were not found

DEAD (= 3)
The group is already dead.

No other fields.

PENDING (= 4)
The group is pending.

err_info

string

informing message that the group is still pending

request
GSGroupRemoveFrom

implementation(s): Python

GSGroupCreate

type enum
GS_GROUP_CREATE

purpose
Ask Global Services to create a group of resources.

fields

user_name

string

user supplied name for the group

items

list[tuple[int, dragon.infrastructure.messages.Message]]

list of tuples where each tuple contains a replication factor and the Dragon
create message

policy

infrastructure.policy.Policy or dict

policy object that controls the placement of the resources

response
GSGroupCreateResponse

implementation(s): Python

GSGroupCreateResponse

type enum
GS_GROUP_CREATE_RESPONSE

purpose
Response to GSGroupCreate message

fields

desc

base64 encoded string

The group descriptor

request
GSGroupCreate

implementation(s): Python

GSGroupKill

type enum
GS_GROUP_KILL

purpose
Ask Global Services to send the processes belonging to a specified group a specified signal.

fields

g_uid

integer

group UID

user_name

string

user supplied name for the group

sig

int

signal to use to kill the process, default=signal.SIGKILL

response
GSGroupKillResponse

implementation(s): Python

GSGroupKillResponse

type enum
GS_GROUP_KILL_RESPONSE

purpose
Response to GSGroupKill message

fields
Alternatives on err:

SUCCESS (= 0)
Group has successfully sent the kill requests.

desc

base64 encoded string

The group descriptor

UNKNOWN (= 1)
No such group is known.

err_info

string

explanation of what went wrong

DEAD (= 2)
The group is already dead.

No other fields.

PENDING (= 3)
The group is pending.

err_info

string

informing message that the group is still pending

ALREADY (= 4)
All the processes of the group were already dead.

desc

base64 encoded string

The group descriptor

request
GSGroupKill

implementation(s): Python

GSGroupCreateAddTo

type enum
GS_GROUP_CREATE_ADD_TO

purpose
Ask Global Services to create and add resources to an existing group of resources.

fields

g_uid

integer

group UID

user_name

string

user supplied name for the group

items

list of strings or integers

list of user supplied names or UIDs for the resources

policy

policy object that controls the placement of the resources

response
GSGroupCreateAddToResponse

implementation(s): Python

GSGroupCreateAddToResponse

type enum
GS_GROUP_CREATE_ADD_TO_RESPONSE

purpose
Response to GSGroupCreateAddTo message

fields
Alternatives on err:

SUCCESS (= 0)
Group has been found and the resources have been created and added to it.

desc

base64 encoded string

The group descriptor

UNKNOWN (= 1)
No such group is known.

err_info

string

explanation of what went wrong

DEAD (= 2)
The group is already dead.

No other fields.

PENDING (= 3)
The group is pending.

err_info

string

informing message that the group is still pending

request
GSGroupCreateAddTo

implementation(s): Python

GSGroupDestroyRemoveFrom

type enum
GS_GROUP_DESTROY_REMOVE_FROM

purpose
Ask Global Services to destroy and remove specific resources from an existing group of resources.

fields

g_uid

integer

group UID

user_name

string

user supplied name for the group

items

list of strings or integers

list of user supplied names or UIDs for the resources

response
GSGroupDestroyRemoveFromResponse

implementation(s): Python

GSGroupDestroyRemoveFromResponse

type enum
GS_GROUP_REMOVE_FROM_RESPONSE

purpose
Response to GSGroupDestroyRemoveFrom message

fields
Alternatives on err:

SUCCESS (= 0)
Group has been found and the resources have been removed from it.

desc

base64 encoded string

The group descriptor

UNKNOWN (= 1)
No such group is known.

err_info

string

explanation of what went wrong

FAIL (= 2)
The request has failed. This can happen when at least one of the resources to be removed from the group was not found.

err_info

string

informing message with the UIDs of those resources that were not found

DEAD (= 3)
The group is already dead.

No other fields.

PENDING (= 4)
The group is pending.

err_info

string

informing message that the group is still pending

request
GSGroupDestroyRemoveFrom

implementation(s): Python

GSGroupList

type enum
GS_GROUP_LIST

purpose
Request a list of the g_uid for all the groups of resources currently being managed

fields
None additional

response
GSGroupListResponse

see also
GSGroupQuery

refer to the Common Fields section for additional request message fields

implementation(s): Python

GSGroupListResponse

type enum
GS_GROUP_LIST_RESPONSE

purpose
Response to GSGroupList message

fields

glist

list of nonnegative integers for all the groups

request
GSGroupList

implementation(s): Python

GSGroupQuery

type enum
GS_GROUP_QUERY

purpose
Request the GroupDescriptor for a managed group of resources

fields
One of the g_uid and user_name fields must be present.

If both are specified, then the user_name field is ignored.

g_uid

integer

target group UID

user_name

string

user supplied name for the target group

response
GSGroupQueryResponse

refer to the Common Fields section for additional request message fields

implementation(s): Python

GSGroupQueryResponse

type enum
GS_GROUP_QUERY_RESPONSE

purpose
Response to request for GroupDescriptor for a managed group

fields
Alternatives on err:

SUCCESS (= 0)
Group has been found

desc

map

initializer for GroupDescriptor - includes at a minimum g_uid of new group and assigned user name.

UNKNOWN (= 1)
No such group is known.

err_info

string

explanation of what went wrong

request
GSGroupQuery

see also
GSGroupList

implementation(s): Python

Other Messages

These messages are related to other transactions on Global Services, for example ones related to sequencing runtime startup and teardown.

AbnormalTermination
type enum
ABNORMAL_TERMINATION

purpose
Error result for startup and teardown messages, as well as for reporting problems that happen during execution that are not tied to any specific request.

Morally speaking, this message backs an exception class related to things going wrong in Dragon Run-time Services. These are generally nonrecoverable errors, but it is desirable to propagate an explanation of what went wrong, back to the front end when possible - users won’t want to collect disparate log files for basic bug reports.

fields
err_info
string

explanation of what went wrong.
implementation(s): Python

GSStarted

type enum
GS_STARTED

purpose
Confirm to Launcher that the Global Services process (and if applicable, subsidiary processes) are started.

fields
None

implementation(s): Python

GSPingSH

type enum
GS_PING_SH

purpose
Confirm to Shepherd(s) that Global Services has started and request handshake response.

fields
None

see also
SHPingGS

implementation(s): Python

GSIsUp

type enum
GS_IS_UP

purpose
Confirm to Launcher that Global Services is completely up

fields
None

implementation(s): Python

GSHeadExit
type enum
GS_HEAD_EXIT

purpose
Notify Launcher that the head process has exited. At this point the Launcher can start teardown or start a new head process.

fields
exit_code
integer

exit code the process left with
implementation(s): Python

GSChannelRelease

type enum
GS_CHANNEL_RELEASE

purpose
Tell the Shepherd(s) that Global Services is exiting and will no longer be sending anything through Channels

fields
None

implementation(s): Python

GSHalted

type enum
GS_HALTED

purpose
Notify Launcher that Global Services is halted. The Global Services processes are expected to exit after sending this message, having done everything possible to clean up.

fields
None

implementation(s): Python

GSTeardown

type enum
GS_TEARDOWN

purpose
Direct Global Services to do a clean exit - clean up any remaining global structures and its own sub-workers, if any, and detach from channels.

This should only be given when there isn’t an active head process. If there is such a thing, we should probably take everything down anyway but complain about it.

implementation(s): Python

GSPingProc

type enum
GS_PING_PROC

purpose
When a new managed process wanting to use Global Services comes up, it must wait to receive this message on its Global Services return channel before sending any messages to Global Services.

This message also carries ‘argument’ data, which is defined as binary data to be delivered to the target at startup. This can be used in whatever way the target would like. For example, the implementation of the Python Process object uses this to deliver the Python target function and arguments.

Remark. Even if no arguments are needed, this message must be sent to solve a race with the local Shepherd’s notification to Global Services that the process was successfully started.

fields
mode
integer

explains how the argdata field should be interpreted

Alternatives:

NONE (=0) No arguments are carried

PYTHON_IMMEDIATE (=1) The arguments are found in fields of the argdata field.

PYTHON_CHANNEL (=2) The argdata has fields for a serialized channel to attach to and get the args from otherwise challenge the available space in the pool the argument delivery channel is found in.
argdata
base64 encoded string, fixme, with the binary argument data to be delivered, or nil.
implementation(s): Python

GSDumpState
type enum
GS_DUMP_STATE

purpose
Primarily debugging. Makes global services dump its state in human readable form to a file specified in the message.

Remark. May in the future try for a formal serialization of GS state once more of the design is concrete. May also have a reply to this message.

fields
filename
string

file to open and write the dump to
implementation(s): Python

GSUnexpected
type enum
GS_UNEXPECTED

purpose
Whenever GS gets a message that is not expected as an input to GS (which can depend on the state it is in) and a valid reply handle can be found corresponding to that message, this message gets returned as a response.

If a valid reply handle can’t be found, then GS will log an error but continue to run.

This kind of thing can happen when issuing interactive commands to a global services process from the launcher. If the head process leaves unexpectedly there can be a race; this message allows handling this without putting GS into error.

TODO: put this in GS documentation and xref, post GS doc update.

fields
ref
integer

matches the tag value of the unexpected message, if it has one.
implementation(s): Python

ExceptionlessAbort

type enum
EXCEPTIONLESS_ABORT

purpose
It can be benefical to pass a message through the services runtime that communicates an abnormal termination has been encountered without immediatesly raising an exception. This message can thus be used to coordinate tearing down of resources without immediately raising an exception.

fields
None other than default

implementation(s): Python

Local Services Messages API

These messages are the ones underlying the Local Services API and are sent only to and from the Shepherd. Like all request messages, these Shepherd request messages have a p_uid field denoting the process that sent the request message and r_c_uid field denoting the channel ID the response should be returned to.

LS Process Messages

SHProcessCreate
type enum
SH_PROCESS_CREATE

purpose
Request to Shepherd to launch a process locally.

fields
t_p_uid
integer

process UID of the target process - Global Services will use this to refer to the process in other operations.
exe
string

executable name
args
list of strings

arguments to the executable. [exe, args] == argv
env
map, key=string value=string

Environment variables to be exported before running.

Note that this is additive/override to the shepherd environment - may need to be more fancy here if it turns out we need to add to paths
rundir
string, absolute path

Indicates where the process is to be run from. If empty, does not matter.
options

output_c_uid is the channel id where output, both stdout
and stderr, should be forwarded if present in an SHFwdOutput message. If this field is not present in the options, then output will be forwarded to the Backend. The SHFwdOutput message indicates whether it was from stdout or stderr.
response
SHProcessCreateResponse

see also
SHProcessKill

refer to the Common Fields section for additional request message fields
implementation(s): Python

SHProcessCreateResponse
type enum
SH_PROCESS_CREATE_RESPONSE

purpose
Response to process creation request.

fields
Alternatives on err:
SUCCESS (= 0)
no fields

FAIL (= 1)
err_info

string

what went wrong
request
SHProcessCreate

see also
SHProcessKill
implementation(s): Python

SHMultiProcessKill

type enum
SH_MULTI_PROCESS_KILL

purpose
Request to Shepherd to efficiently kill multiple processes.

fields

procs

list of SHProcessKill messages representing the processes to be killed.

response
SHMultiProcessKillResponse

see also
SHProcessKill

refer to the Common Fields section for additional request message fields

implementation(s): Python

SHMultiProcessKillResponse

type enum
SH_MULTI_PROCESS_KILL_RESPONSE

purpose
Response to multi process kill request.

fields
Alternatives on err:

responses

list of SHProcessKillResponse messages representing the results of each SHProcessKill request.

SUCCESS (= 0)

responses

list of SHProcessKillResponse messages for each previously requested process.

FAIL (= 1)

err_info

string

what went wrong

request
SHMultiProcessKill

see also
SHProcessKill

implementation(s): Python

SHProcessKill
type enum
SH_PROCESS_KILL

purpose
Request to kill a process owned by this shepherd, with a specified signal.

fields
t_p_uid
integer

process UID of the target process
sig
integer, valid signal number

signal to send to the process
response
SHProcessExit Note difference in nomenclature.

see also
SHProcessCreate

refer to the Common Fields section for additional request message fields
implementation(s): Python

SHMultiProcessCreate

type enum
SH_MULTI_PROCESS_CREATE

purpose
Request to Shepherd to launch multiple processes locally.

fields

pmi_group_info

Optional PMIGroupInfo structure.

Contains common PMI/MPI values needed to start the requested PMI enabled applications, including the nid_list, host_list.

procs

list of SHProcessCreate messages representing the processes to be started.

response
SHMultiProcessCreateResponse

see also
SHProcessKill

refer to the Common Fields section for additional request message fields

implementation(s): Python

SHMultiProcessCreateResponse

type enum
SH_MULTI_PROCESS_CREATE_RESPONSE

purpose
Response to multi process creation request.

fields
Alternatives on err:

SUCCESS (= 0)

responses

list of SHProcessCreateResponse messages for each previously requested process.

FAIL (= 1)

err_info

string

what went wrong

request
SHMultiProcessCreate

see also
SHProcessKill

implementation(s): Python

SHProcessKillResponse
type enum
SH_PROCESS_KILL_RESPONSE

purpose
Response to request to kill a process owned by this shepherd. Returns confirmation that the signal was delivered only. The process’s exit comes automatically via SHProcessExit.

fields
Alternatives on err:
SUCCESS (= 0)
no fields

FAIL (= 1)
err_info

string

what went wrong
response
SHProcessKill

see also
SHProcessCreate SHProcessExit

refer to the Common Fields section for additional request message fields
implementation(s): Python

SHProcessExit
type enum
SH_PROCESS_EXIT

purpose
This message is sent to Global Services when a managed process exits. There are no ref or err fields.

*fields
exit_code
int

numerical exit code.
p_uid
int

the Process ID of the process that exited. When a process exits of its own accord, this is the

only thing to identify which one it is.
see also
SHProcessKill
implementation(s): Python

Pool Messages

These messages are directives to a Local Services process to execute pool lifecycle commands. They are posed in the form of individual messages instead of generic operation completions because the Shepherd holds these OS resources.

SHPoolCreate
type enum
SH_POOL_CREATE

purpose
Create a new memory pool.

fields
m_uid
positive integer

system wide unique identifier for this pool. All future operations on this pool will refer to it using the m_uid.
name
string

name of the pool to use when creating it. This name is guaranteed to be globally unique.
size
positive integer

size in bytes of the pool.
attr
string

this string is the base64 encoding of a serialized pool attribute object to use when creating the pool
response
SHPoolCreateResponse

see also
refer to the Common Fields section for additional request message fields
implementation(s): Python

SHPoolCreateResponse
type enum
SH_POOL_CREATE_RESPONSE

purpose
Response to request to create a new memory pool.

fields
Alternatives on err:
SUCCESS (= 0)
The pool was created successfully.

desc

string

the string is the base64 encoding of a serialized pool descriptor for the pool that just got created
FAIL (= 1)
The pool was not created successfully. This could be because the pool creation call failed or because there was something wrong with the pool creation message.

err_info

string

what went wrong
request
SHPoolCreate
implementation(s): Python

SHPoolDestroy

type enum
SH_POOL_DESTROY

purpose
Request to destroy a memory pool.

fields

m_uid

positive integer

ID of the pool to destroy

response
SHPoolDestroyResponse

see also
refer to the Common Fields section for additional request message fields

implementation(s): Python

SHPoolDestroyResponse

type enum
SH_POOL_DESTROY_RESPONSE

purpose
Response to request to destroy a memory pool

fields
Alternatives on err:

SUCCESS (= 0)
The pool was destroyed. No other fields

FAIL (= 1)
The pool was not destroyed successfully. This could be because the pool destroy call failed or because there was something wrong with the pool destroy message.

err_info

string

what went wrong

request
SHPoolDestroy

implementation(s): Python

SHExecMemRequest

type enum
SH_EXEC_MEM_REQUEST

purpose
Request to execute a memory request. This message contains a serialized operation given to the Shepherd to execute on behalf of the requesting process using dragon_memory_exec_req or dragon_memory_pool_exec_request

fields

kind

integer, legal values are 0 or 1

If 0, the request is a serialized dragonMemoryPoolRequest

If 1, the request is a serialized dragonMemoryRequest

request

string

this is the base64 encoding of the request.

response
SHExecMemResponse

see also
refer to the Common Fields section for additional request message fields

implementation(s): Python

SHExecMemResponse

type enum
SH_EXEC_MEM_RESPONSE

purpose
Response to request to execute a memory request. Note that there is no need for this message to include what kind of request it was - the sender knows this.

fields
Alternatives on err:

SUCCESS (= 0)
The memory request call completed. Even if the call did not complete with a success return value, the response has the details about what went wrong including the error code and this will be apparent when the original requester calls the completion function on the response.

response

string

this is the base64 encoding of the serialized response

FAIL (= 1)
The request could not be completed.

err_info

string

what went wrong

err_code

integer

if the error is due to a library call, this should carry the return value of the call.

Otherwise its value should be 0.

implementation(s): Python

Channel Messages

SHChannelCreate
type enum
SH_CHANNEL_CREATE

purpose
Request to create a channel in a memory pool known to this shepherd.

fields
m_uid
integer

Unique id of the pool to create the channel inside.
c_uid
integer

Unique id of the channel to be created. Future operations will use this number to refer to the channel
options
map

Initializer for a (local) ChannelOptions object. This is an open type intended to completely describe the controllable parameters for channel creation at the library level. Will contain attributes string from library.
response
SHChannelCreateResponse

see also
refer to the Common Fields section for additional request message fields

implementation(s): Python
Python

SHChannelCreateResponse
type enum
SH_CHANNEL_CREATE_RESPONSE

purpose
Response to channel allocation request.

fields
Alternatives on err:
SUCCESS (= 0)
desc

string

serialized and ascii encoded library channel descriptor
FAIL (= 1)
err_info

string

what went wrong
request
SHChannelCreate
implementation(s): Python

SHChannelDestroy
type enum
SH_CHANNEL_DESTROY

purpose
Request to free a previously allocated channel.

fields
c_uid
integer

unique id of the channel to be destroyed.
response
SHChannelDestroyResponse

see also
refer to the Common Fields section for additional request message fields
implementation(s): Python

SHChannelDestroyResponse
type enum
SH_CHANNEL_DESTROY_RESPONSE

purpose
Response to request to free a previously allocated channel.

fields
Alternatives on err:
SUCCESS (= 0)
No additional fields

FAIL (= 1)
err_info

string

what went wrong
request
SHChannelDestroy
implementation(s): Python

SHLockChannel

PLACEHOLDER
type enum
SH_LOCK_CHANNEL

purpose
Request to lock a channel

TODO: should only something attached to a channel be allowed to lock it?

fields
read_lock
bool

locking this for reading
write_lock
bool

locking this for writing
response

SHLockChannelResponse

see also
refer to the Common Fields section for additional request message fields
implementation(s): Python

SHLockChannelResponse

PLACEHOLDER

type enum
SH_LOCK_CHANNEL_RESPONSE

purpose
Response to request to lock a channel

fields
Alternatives on err:

SUCCESS (= 0)
Successfully locked.

ALREADY (= 1)
Some other entity has already locked the channel.

request
SHLockChannel

see also
x

implementation(s): Python

Memory Allocation Messages

SHAllocMsg

PLACEHOLDER - maybe OBE

type enum
SH_ALLOC_MSG

purpose
Request a shared memory allocation for a large message

fields
x

response
SHAllocMsgResponse

see also
refer to the Common Fields section for additional request message fields

implementation(s): Python

SHAllocMsgResponse

PLACEHOLDER - OBE?

type enum
SH_ALLOC_MSG_RESPONSE

purpose
Response to a requested allocation for a large message

fields
Alternatives on err:

SUCCESS (= 0)

request
SHAllocMsg

see also
x

implementation(s): Python

SHAllocBlock

PLACEHOLDER - OBE?
type enum
SH_ALLOC_BLOCK

purpose
Request a shared memory allocation for generic memory

fields
p_uid
integer

Process UID of the requesting process.

Note that the Shepherd may need to know this to do cleanup if the process goes away. There probably needs to be an option as to which is the owning process.
len
integer

length of allocation desired
response
SHAllocBlockResponse

see also
refer to the Common Fields section for additional request message fields
implementation(s): Python

SHAllocBlockResponse

PLACEHOLDER - OBE?
type enum
SH_ALLOC_BLOCK_RESPONSE

purpose
Response to a requested allocation for generic memory

fields
Alternatives on err:
SUCCESS (= 0)
Successfully allocated the block. Do we want a reference ID for the block to let other processes claim ownership? E.g. we could have the shepherd count ref on such blocks.

seg_name

string

name of shared memory segment the block is found in.

offset

nonnegative integer

byte offset of the allocated block within segment
UNKNOWN (= 1)
The p_uid of the requester is not a process currently running and being managed by the shepherd

FAIL (= 2)
The block could not be allocated.

err_info

string

what went wrong
request
SHAllocBlock

see also
x
implementation(s): Python

Other Messages

These Local Services messages are for other purposes, such as sequencing bringup and teardown. Where appropriate they have the idx field, indicating which Shepherd instance there is. This is an integer inside [0,N), where N is the number of compute nodes in the allocation. It has the value 0 in the single node case.

SHChannelsUp
type enum
SH_CHANNELS_UP

purpose
Notify Launcher that this Shepherd has allocated the shared memory for the channels and initialized whatever default channels there are.

fields
node_desc
Python<dragon.infrastructure.node_desc.NodeDescriptor or its serialized descriptor
idx
integer

which Shepherd in the allocation this is. Index is 0 in the single node case and lives in [0,N) for N multi node back end allocation
gs_cd
base64 encoded string

When idx is the primary node, then this is the channel descriptor of global services. Otherwise it is ignored and should be set to an empty string.
implementation(s): Python
Python

SHPingGS
type enum
SH_PING_GS

purpose
Acknowledge to Global Services that this Shepherd is up and ready.

fields
idx
integer

which Shepherd in the allocation this is. Index is 0 in the single node case and lives in [0,N) for N multi node back end allocation.
node_sdesc
dict

serialized NodeDescriptor

of the node the Shepherd is running on. - Note that the h_uid in the descriptor will be None in this message. It is set later by GS.
implementation(s): Python

SHHalted
type enum
SH_HALTED

purpose
Notify launcher that this Shepherd is halted.

fields
idx
integer

which Shepherd in the allocation this is. Index is 0 in the single node case and lives in [0,N) for N multi node back end allocation
implementation(s): Python

SHTeardown

type enum
SH_TEARDOWN

purpose
Direct Shepherd to do a clean teardown.

fields
None other than default

implementation(s): Python

SHPingBE
type enum
SH_PING_BE

purpose
Shepherd handshake with MRNet backend

fields
shep_cd
Channel Descriptor for the Shepherd (serialized, base64 encoded)
be_cd
The Launcher Back End Channel Descriptor (serialized, base64 encoded)
gs_cd
The Global Services Channel Descriptor (serialized, base64 encoded)
default_pd
The Default Pool Descriptor (serialized, base64 encoded)
inf_pd
The Infrastructure Pool Descriptor (serialized, base64 encoded)
implementation(s): Python

SHHaltBE

type enum
SH_HALT_BE

purpose
Shepherd telling the MRNet backend to exit.

fields
None other than default

implementation(s): Python

SHFwdInput
type enum
SH_FWD_INPUT

purpose
Message carrying data intended to be written into the standard input of a process managed by this shepherd.

If the process is present and the stdin can be written without any problems, there is no response message.

fields
t_p_uid
integer

target p_uid
input
string

input to be sent to target process

The max length of any input to be sent through this mechanism in one message is 1024 bytes.
confirm
boolean

request confirmation of forwarded input.
response
SHFwdInputErr
implementation(s): Python

SHFwdInputErr

type enum
SH_FWD_INPUT

purpose
Error response to a forward input message. This message is returned to the sender if (for instance) the target process isn’t present at the time of delivery.

fields
All messages:
idx
integer

which Shepherd in the allocation this is. Index is 0 in the single node case and lives in [0,N) for N multi node back end allocation
Alternatives on err:

SUCCESS (= 0)
No fields other than defaults.

No error. This message isn’t expected ever to be emitted, but is called out here to reserve the ‘0’ code for future use in case there needs to be positive confirmation of message delivery.

FAIL (= 1)
err_info

string

explanation of what went wrong.
request
SHFwdInput

implementation(s): Python

SHFwdOutput

type enum
SH_FWD_OUTPUT

purpose
Message carrying data from either stdout or stderr of a process managed by this shepherd. This comes from the Shepherd and no response is expected.

fields
fdnum

Indicates which stream, stdout or stderr, is being forwarded.

Alternatives for fdnum:

stdout (= 1) stderr (= 2)
data
string

The data read from the stream and being forwarded

The max length of any output per message is 1024 bytes.
idx
integer

which Shepherd in the allocation this is. Index is 0 in the single node case and lives in [0,N) for N multi node back end allocation
p_uid
integer

The process UID this came from
implementation(s): Python

SHHaltTA

type enum
SH_HALT_TA

purpose
Message coming from Launcher to the Shepherd, telling it to tell TA to halt. The Shepherd forwards this same message to TA.

fields
None other than default

implementation(s): Python

SHDumpState
type enum
SH_DUMP_STATE

purpose
Primarily debugging. Makes the Shepherd dump its state in human readable form to a file specified in the message or dump to the log if no filename is provided.

fields
filename
string

file to open and write the dump to. Defaults to None. If None, then the dump goes to the log file for the Shepherd.
implementation(s): Python

Launcher Messages API

These messages go to the Launcher frontend in standard and server mode via the LA Channel.

LABroadcast

type enum

LA_BROADCAST

purpose
This message is sent by the launcher to the Launch Router. This message contains another message in the data field. The message in the data field is to be broadcast to the Shepherd on all nodes through the Backend. The LABroadcast message is sent to the backend. The Backend extracts the embedded message from the LABroadcast message and sets the embedded message’s r_c_uid as the return channel id in the message and forwards the embedded message to the Shepherd’s channel. All broadcast messages are sent to the Shepherd on each node.

fields

data
A string to be broadcast to all nodes. For the Dragon infrastructure this data field will itself be a valid, serialized message.
implementation(s): Python, Network Front End has a dependency on the type enum being 68.

LAPassThruFB

type enum

LA_PASS_THRU_FB

purpose

This message is a pass thru message from the Launcher to the Backend.

When sent to the Launcher Router, it forwards the message to the Backend on the primary node.

This message type is used by the Dragon Kernel Server to communicate its channel id to the Dragon Kernel indicating that the Dragon Kernel is up and ready to accept requests.

fields
c_uid
The channel to send this message to running on a compute node.
data
A user-defined serialized message that is unwrapped by the receiving end.
NOTE
The r_c_uid is set in this message by the Backend to the Backend channel id before it is forwarded on to its destination on the compute node.
implementation(s): Python

LAPassThruBF

type enum

LA_PASS_THRU_BF

purpose

This message is a pass thru message from the Backend to the Launcher.

Note: One use of this message type is by a server back end to communicate its channel id to a server front end in the r_c_uid field. This serves as an indicator in this case that a front end that its corresponding back end is ready to accept requests (see Launcher Server Mode).

fields
data
A user-defined serialized message that is unwrapped by the

receiving end.
NOTE
The r_c_uid is set to the channel id of the source of the

passthrough message on a compute node.
implementation(s): Python

LAServerMode

type enum

LA_SERVER_MODE

purpose

This is a message that is used internally in the Launcher to set up Server Mode. It is not ever to be sent between components in the Dragon run-time.

fields
frontend
A string with a file specification for the server front end to be run on the login node.
backend
A string with a file specification for the server back end to be run on the primary compute node.
frontendargs
A list of strings containing any command-line arguments to the front end.
backendargs
A list of strings containing any command-line arguments to the back end.
implementation(s): Python

LAServerModeExit

type enum

LA_SERVER_MODE_EXIT

purpose

This is a message that is used internally in the Launcher to exit Server Mode. It is not ever to be sent between components in the Dragon run-time.

fields

None other than default

implementation(s): Python

LAProcessDict

type enum
LA_PROCESS_DICT

purpose
Return a dictionary of process information for all the processes that were started in the launcher via the run command.

fields
None additional

response
LAProcessDictResponse

see also
refer to the Common Fields section for additional request message fields

implementation(s): Python

LAProcessDictResponse
type enum
LA_PROCESS_DICT_RESPONSE

purpose
Responds with a dictionary for all the processes were started in the launcher via the run command. The dictionary includes fields for exe, args, env, run directory, user name, options, create return code, and exit code.

fields
pdict
dictionary of p_uid to process information as described.
request
LAProcessList
implementation(s): Python

LADumpState
type enum
LA_DUMP_STATE

purpose
Primarily debugging. Makes the Launcher dump its state in human readable form to a file specified in the message or dump to the log if no filename is provided.

fields
filename
string

file to open and write the dump to. Defaults to None. If None, then the dump goes to the log file for the Launcher.
implementation(s): Python

LANodeIdxSH
type enum
LA_NODEINDEX_SH

purpose
Communicates the node index from the Launcher Back End to the Shepherd during bootstrapping of the environment.

fields
node_idx
integer

The index of this node in the allocation upon which the Dragon runtime services are running.
implementation(s): Python

LAChannelsInfo
type_enum
LA_CHANNELS_INFO

purpose
Broadcast to all nodes to provide hostnames, node indices, and channel descriptors to all nodes in the allocation so services like global services and the inter-node transport service (HSTA, TCP or otherwise) can be started.

When the launcher backend receives this message, it forwards it to the local shepherd. The local shepherd provides this message to the inter-node transport service when it is started and to global services (on the primary node) when it is started.

fields
nodes_desc
dictionary with keys corresponding to string node indices and values of class

dragon.infrastructure.node_desc.NodeDescriptor``(eg: The hostname info for node 5 is accessed via ``la_channels_msg.nodes_desc['4'].host_name ) - Excamples attributes for each value of dragon.infrastructure.node_desc.NodeDescriptor:

‘host_name’: hostname string

‘host_id’: integer which is determined by calling the Posix gethostid function

‘ip_addr’: ip address string

‘shep_cd’: base64 encoded shepherd channel descriptor
gs_cd
string

The base64 encoded channel descriptor of global services
num_gw_channels
int

The number of gateway channels per node
implementation(s): Python
Python

Breakpoint
type_enum
BREAKPOINT

purpose
Inform front end that a managed process has reached a breakpoint for the first time and wants to establish a debug connection.

fields
p_uid

int

process uid that has just reached a breakpoint

index

int

node that process is on

out_desc

string

encoded binary serialized channel descriptor for debug connection outbound

in_desc

string

encoded binary serialized channel descriptor for debug connection inbound
implementation(s): Python

Launcher Backend Messages APIs

These messages are used by a combination of the Launcher and the Launcher Backend to support the launcher in both regular and server mode. They are communicated through the BELA Channel.

BEIsUp

type enum
BE_IS_UP

purpose
Confirm to Launcher Frontend that the Backend is up and send its serialized channel descriptor and host_id. This message is not communicated through the BELA Channel, as this is not set up yet.

fields

be_ch_desc

Backend serialized channel descriptor. The frontend will attach to

this channel and establish a connection with the backend.

host_id

integer

the hostid of the node sending this message.

implementation(s): Python

BEPingSH

type enum
BE_PING_SH

purpose
MRNet backend handshake with Shepherd

fields
None other than default

implementation(s): Python

BEHalted

type enum
BE_HALTED

purpose
Indicate that the MRNet backend instance on this node has exited normally.

fields

None other than default

implementation(s): Python

BENodeIdxSH

type enum
BE_NODE_IDX_SH

purpose

This message is sent from the Launcher Backend to Local Services during initialization to tell Local Services its node index and related host info for the current run. All nodes are numbered 0 to n-1 where n is the number of nodes in the allocation.

fields
node_idx
An integer representing the node index value for this node.
host_name
string

name of host as returned by MRNet backend
ip_addr
string

IP address as returned by DNS lookup of hostname
primary
string

hostname of primary node
logger_sdesc
dragon.utils.B64

serialized descriptor of backend logging channel
implementation(s): Python

Launcher Frontend Messages APIs

These messages are used by a combination of the Launcher and the Launcher Frontend to support the launcher in both regular and server mode.

FENodeIdxBE

type enum
FE_NODE_IDX_BE

purpose
The Frontend sends the node index to the Backend, based on the backend’s host_id.

fields

node_index

integer

The index of this node in the allocation upon which the Dragon runtime services are running.

implementation(s): Python

HaltOverlay

type enum
HALT_OVERLAY

purpose
Indicate that monitoring of the overlay network should cease

fields
None other than default

implementation(s): Python

HaltLoggingInfra

type enum
HALT_LOGGING_INFRA

purpose
Indicate that monitoring of logging messages from the backend should cease

fields
None other than default

implementation(s): Python

LAExit

type enum
LA_EXIT

purpose
Indicate the launcher should exit. Use in case the launcher teardown was unable to complete correctly.

fields
None other than default

implementation(s): Python

Logging Messages API

In order to transmit messages from backend services to the launcher frontend, the logging infrastructure packs logging strings into messages that are passed from a given backend service to the MRNet server backend, which aggregates any logging messages its received and sends them to the launcher frontend to be offically logged to file or terminal.

LoggingMsg

type enum
LOGGING_MSG

purpose
To take logging strings provided via python logging (eg: log.info('message')) and pack all information that’s necessary to make log messages useful on the frontend

fields
All fields are auto-populated by the logging infrastructure found in LoggingInfrastructure
name
string

name of logger message came from. Typically formatted f{service}.{func}
msg
string

log mesage
time
string

time generated by the sending logger’s formatter
func
string

name of function the logging msg was made in
hostname
string

hostname of sender
ip_address
string

IP address of sender
port
string

Port of sender (placeholder for now but likely useful in single-node testing)
service
string

service as defined in Python
level
int

logging level as defined by dragonMemoryPoolAttr_t and is consistent with Python’s only loggin int values
implementation(s) Python

LoggingMsgList

type enum
LOGGING_MSG_LIST

purpose
Takes a list of LoggingMsg and aggregates them into a single message the MRNet server backend can send up to the MRNet frontend server

fields

records
list of LoggingMsg messages with attributes as defined therein

implementation(s) Python

LogFlushed

type enum
LOG_FLUSHED

purpose
Sent by MRNet server backend to frontend after it has completed its final flushing and transmission of logs to the fronted. Part of the transaction diagram for teardown in Multi Node Teardown

fields
None other than default

implementation(s) Python

Overlay/Transport Agent Messages API

In multi node cases, one or more transport agent processes live on each node and they need a few messages related to setup and teardown control through the TA Channel.

TAPingSH

type enum
TA_PING_SH

purpose
Indicate that the TA instance on this node has come up and succeeded in getting communication with the other TA instances.

fields
None other than default

implementation(s): Python

TAHalted
type enum
TA_HALTED

purpose
Indicate that the TA instance on this node has exited normally.

fields
idx
integer

which node in the allocation this is.
implementation(s): Python

TAUp

type enum
TA_UP

purpose
Indicate that the TA instance on this node is up and ready.

fields
None other than default

implementation(s): Python

OverlayPingBE

type enum
OVERLAY_PING_BE

purpose
Indicate that the Overlay instance on this backend node is up and ready.

fields
None other than default

implementation(s): Python

OverlayPingLA

type enum
OVERLAY_PING_LA

purpose
Indicate that the Overlay instance on this frontend node is up and ready.

fields
None other than default

implementation(s): Python

LAHaltOverlay

type enum
LA_HALT_OVERLAY

purpose]
Launcher frontend requests a shutdown of Overlay agent

fields
None other than default

implementation(s): Python

BEHaltOverlay

type enum
BE_HALT_OVERLAY

purpose]
This backend node requests a shutdown of its Overlay agent

fields
None other than default

implementation(s): Python

OverlayHalted

type enum
OVERLAY_HALTED

purpose]
This overlay instance has shutdown

fields
None other than default

implementation(s): Python

Python Infrastructure Messages Implementation

The following subsections contain structures relating to the Python implementation of the Dragon infrastructure message definitions. The specification of these message types i`` in the previous section and cross-referenced from these two sections for convenience.

Python Implementation of Infrastructure Messages

Alphabetical listing of all message classes in the Python implementation of the infrastructure messages.

Dragon infrastructure messages are the internal API used for service communication.

class FileDescriptor

Bases: Enum

An enumeration.

stdin = 0

stdout = 1

stderr = 2

exception AbnormalTerminationError

Bases: Exception

__init__(msg='')

class PMIGroupInfo

Bases: object

Required information to enable the launching of pmi based applications.

job_id: int

nnodes: int

nranks: int

nidlist: list[int]

hostlist: list[str]

control_port: int

classmethod fromdict(d)

__init__(job_id: int, nnodes: int, nranks: int, nidlist: list[int], hostlist: list[str], control_port: int) → None

class PMIProcessInfo

Bases: object

Required information to enable the launching of pmi based applications.

lrank: int

ppn: int

nid: int

pid_base: int

classmethod fromdict(d)

__init__(lrank: int, ppn: int, nid: int, pid_base: int) → None

class InfraMsg

Bases: object

Common base for all messages.

This common base type for all messages sets up the default fields and the serialization strategy for now.

class Errors

Bases: Enum

An enumeration.

INVALID = -1

__init__(tag, ref=None, err=None)

get_sdict()

property tc

classmethod tcv()

property tag

property ref

property err

classmethod from_sdict(sdict)

classmethod deserialize(msg)

uncompressed_serialize()

serialize()

class CapNProtoMsg

Bases: object

Common base for all capnproto messages.

This common base type for all messages sets up the default fields and the serialization strategy for messages to be exchanged between C and Python.

Errors: alias of DragonError

__init__(tag)

classmethod from_sdict(sdict)

classmethod deserialize(msg_str)

serialize()

get_sdict()

builder()

property capnp_name

property tc

property tag

class CapNProtoResponseMsg

Bases: CapNProtoMsg

Common base for all capnproto response messages.

This provides some support for code common to all response messages.

__init__(tag, ref, err, errInfo)

get_sdict()

builder()

property ref

property err

property errInfo

class SHCreateProcessLocalChannel

Bases: CapNProtoMsg

__init__(tag, puid, respFLI)

get_sdict()

builder()

property respFLI

property puid

class SHCreateProcessLocalChannelResponse

Bases: CapNProtoResponseMsg

__init__(tag, ref, err, errInfo='', serChannel='')

get_sdict()

builder()

property serialized_channel

class SHPushKVL

Bases: CapNProtoMsg

__init__(tag, key, value, respFLI)

get_sdict()

builder()

property key

property value

property respFLI

class SHPushKVLResponse

Bases: CapNProtoResponseMsg

__init__(tag, ref, err, errInfo='')

class SHPopKVL

Bases: CapNProtoMsg

__init__(tag, key, value, respFLI)

get_sdict()

builder()

property key

property value

property respFLI

class SHPopKVLResponse

Bases: CapNProtoResponseMsg

__init__(tag, ref, err, errInfo='')

class SHGetKVL

Bases: CapNProtoMsg

__init__(tag, key, respFLI)

get_sdict()

builder()

property key

property respFLI

class SHGetKVLResponse

Bases: CapNProtoResponseMsg

__init__(tag, ref, err, errInfo='', values=[])

get_sdict()

builder()

property values

class SHSetKV

Bases: CapNProtoMsg

__init__(tag, key, value, respFLI)

get_sdict()

builder()

property key

property value

property respFLI

class SHSetKVResponse

Bases: CapNProtoResponseMsg

__init__(tag, ref, err, errInfo='')

class SHGetKV

Bases: CapNProtoMsg

__init__(tag, key, respFLI)

get_sdict()

builder()

property key

property respFLI

class SHGetKVResponse

Bases: CapNProtoResponseMsg

__init__(tag, ref, err, errInfo='', value=None)

get_sdict()

builder()

property value

class DDCreate

Bases: CapNProtoMsg

__init__(tag, respFLI, args)

get_sdict()

builder()

property respFLI

property args

class DDCreateResponse

Bases: CapNProtoResponseMsg

__init__(tag, ref, err, errInfo='')

class DDGetRandomManager

Bases: CapNProtoMsg

__init__(tag, respFLI)

get_sdict()

builder()

property respFLI

class DDGetRandomManagerResponse

Bases: CapNProtoResponseMsg

__init__(tag, ref, err, manager, errInfo='')

get_sdict()

builder()

property manager

class DDRegisterClient

Bases: CapNProtoMsg

__init__(tag, respFLI, bufferedRespFLI)

get_sdict()

builder()

property respFLI

property bufferedRespFLI

class DDRegisterClientResponse

Bases: CapNProtoResponseMsg

__init__(tag, ref, err, clientID, numManagers, managerID, managerNodes, timeout, errInfo='')

get_sdict()

builder()

property clientID

property numManagers

property managerID

property managerNodes

property timeout

class DDConnectToManager

Bases: CapNProtoMsg

__init__(tag, clientID, managerID)

get_sdict()

builder()

property clientID

property managerID

class DDConnectToManagerResponse

Bases: CapNProtoResponseMsg

__init__(tag, ref, err, manager, errInfo='')

get_sdict()

builder()

property manager

class DDDestroy

Bases: CapNProtoMsg

__init__(tag, clientID, respFLI)

get_sdict()

builder()

property respFLI

property clientID

class DDDestroyResponse

Bases: CapNProtoResponseMsg

__init__(tag, ref, err, errInfo='')

class DDRegisterManager

Bases: CapNProtoMsg

__init__(tag, mainFLI, respFLI, hostID)

get_sdict()

builder()

property mainFLI

property respFLI

property hostID

class DDRegisterManagerResponse

Bases: CapNProtoResponseMsg

__init__(tag, ref, err, managerID, errInfo='', managers=[], managerNodes=[])

get_sdict()

builder()

property managerID

property managers

property managerNodes

class DDRegisterClientID

Bases: CapNProtoMsg

__init__(tag, clientID, respFLI, bufferedRespFLI)

get_sdict()

builder()

property clientID

property respFLI

property bufferedRespFLI

class DDRegisterClientIDResponse

Bases: CapNProtoResponseMsg

__init__(tag, ref, err, errInfo='')

class DDDestroyManager

Bases: CapNProtoMsg

__init__(tag, respFLI)

get_sdict()

builder()

property respFLI

class DDDestroyManagerResponse

Bases: CapNProtoResponseMsg

__init__(tag, ref, err, errInfo='')

class DDPut

Bases: CapNProtoMsg

__init__(tag, clientID, chkptID=0, persist=True)

get_sdict()

builder()

property clientID

property chkptID

property persist

class DDPutResponse

Bases: CapNProtoResponseMsg

__init__(tag, ref, err, errInfo='')

class DDGet

Bases: CapNProtoMsg

__init__(tag, clientID, chkptID=0)

get_sdict()

builder()

property clientID

property chkptID

class DDGetResponse

Bases: CapNProtoResponseMsg

__init__(tag, ref, err, errInfo='')

class DDPop

Bases: CapNProtoMsg

__init__(tag, clientID, chkptID=0)

get_sdict()

builder()

property clientID

property chkptID

class DDPopResponse

Bases: CapNProtoResponseMsg

__init__(tag, ref, err, errInfo='')

class DDContains

Bases: CapNProtoMsg

__init__(tag, clientID, chkptID=0)

get_sdict()

builder()

property clientID

property chkptID

class DDContainsResponse

Bases: CapNProtoResponseMsg

__init__(tag, ref, err, errInfo='')

class DDGetLength

Bases: CapNProtoMsg

__init__(tag, clientID, respFLI, chkptID=0)

get_sdict()

builder()

property clientID

property respFLI

property chkptID

class DDGetLengthResponse

Bases: CapNProtoResponseMsg

__init__(tag, ref, err, errInfo='', length=0)

get_sdict()

builder()

property length

class DDClear

Bases: CapNProtoMsg

__init__(tag, clientID, chkptID, respFLI)

get_sdict()

builder()

property clientID

property chkptID

property respFLI

class DDClearResponse

Bases: CapNProtoResponseMsg

__init__(tag, ref, err, errInfo='')

class DDManagerStats

Bases: CapNProtoMsg

__init__(tag, respFLI)

get_sdict()

builder()

property respFLI

class DDManagerStatsResponse

Bases: CapNProtoResponseMsg

__init__(tag, ref, err, errInfo='', data='')

get_sdict()

builder()

property data

class DDManagerGetNewestChkptID

Bases: CapNProtoMsg

__init__(tag, respFLI)

get_sdict()

builder()

property respFLI

class DDManagerGetNewestChkptIDResponse

Bases: CapNProtoResponseMsg

__init__(tag, ref, err, errInfo='', managerID=0, chkptID=0)

get_sdict()

builder()

property chkptID

property managerID

class DDGetIterator

Bases: CapNProtoMsg

__init__(tag, clientID, chkptID=0)

get_sdict()

builder()

property clientID

property chkptID

class DDGetIteratorResponse

Bases: CapNProtoResponseMsg

__init__(tag, ref, err, errInfo='', iterID=0)

get_sdict()

builder()

property iterID

class DDIteratorNext

Bases: CapNProtoMsg

__init__(tag, clientID, iterID)

get_sdict()

builder()

property clientID

property iterID

class DDIteratorNextResponse

Bases: CapNProtoResponseMsg

__init__(tag, ref, err, errInfo='')

class DDKeys

Bases: CapNProtoMsg

__init__(tag, clientID, chkptID=0)

get_sdict()

builder()

property clientID

property chkptID

class DDKeysResponse

Bases: CapNProtoResponseMsg

__init__(tag, ref, err, errInfo='')

class DDDeregisterClient

Bases: CapNProtoMsg

__init__(tag, clientID, respFLI)

get_sdict()

builder()

property clientID

property respFLI

class DDDeregisterClientResponse

Bases: CapNProtoResponseMsg

__init__(tag, ref, err, errInfo='')

class GSProcessCreate

Infrastructure Messages

Generalities about Messages

Common Fields

The _tc TypeCode Field

The tag Field

Message Categories

The ref Field

The err Field

Format of Messages List

Global Services Messages API

GS Process Messages

Pool Messages

Channel Messages

Node Messages

Other Messages

Local Services Messages API

LS Process Messages

Pool Messages

Channel Messages

Memory Allocation Messages

Other Messages

Launcher Messages API

Launcher Backend Messages APIs

Launcher Frontend Messages APIs

Logging Messages API

Overlay/Transport Agent Messages API

Python Infrastructure Messages Implementation

Python Implementation of Infrastructure Messages

Python Message Type Value Enumeration