stormlog.infer.server_group
Declared groups of server processes, such as tensor-parallel GPU workers.
Each collector in a group passes the same --group-id and --world-size
and its own --rank. Membership is declared, never inferred: an identity
without the group ID, a second identity for one rank (a restarted worker or a
changed GPU), or a missing rank keeps the join from happening.
Functions
|
Group samples by identity, ordered by rank for a declared group. |
|
Return why the samples are neither one server nor one complete group. |
- stormlog.infer.server_group.membership_issue(samples)[source]
Return why the samples are neither one server nor one complete group.
- Parameters:
samples (Iterable[TelemetrySample])
- Return type:
str | None
- stormlog.infer.server_group.members(samples)[source]
Group samples by identity, ordered by rank for a declared group.
- Parameters:
samples (Iterable[TelemetrySample])
- Return type:
dict[ServerIdentity, list[TelemetrySample]]