Agent: Change container_ids from list to set in stat collection

[fregataa]

Problem

In src/ai/backend/agent/agent.py, the container_ids collection is built as a list, which could allow duplicate container IDs to be processed multiple times during stat collection.

Affected Code

*File*: src/ai/backend/agent/agent.py

Line 1391 (collect_container_stat)

container_ids: list[ContainerId] = []

for kernel_obj in [\*self.kernel_registry.values()]:

    if not kernel_obj.stats_enabled or kernel_obj.container_id is None:

        continue

    container_ids.append(ContainerId(kernel_obj.container_id))

await self.stat_ctx.collect_container_stat(container_ids)

Line 1403 (collect_process_stat)

container_ids = []

for kernel_obj in [\*self.kernel_registry.values()]:

    if not kernel_obj.stats_enabled or kernel_obj.container_id is None:

        continue

    container_ids.append(ContainerId(kernel_obj.container_id))

await self.stat_ctx.collect_per_container_process_stat(container_ids)

Proposed Solution

Change from list to set to automatically deduplicate container IDs:

container_ids: set[ContainerId] = set()

for kernel_obj in [\*self.kernel_registry.values()]:

    if not kernel_obj.stats_enabled or kernel_obj.container_id is None:

        continue

    container_ids.add(ContainerId(kernel_obj.container_id))

Benefits

1. *Automatic deduplication*: Prevents duplicate stat collection if same container ID appears multiple times

2. *Performance*: Avoids redundant stat collection calls

3. *Semantic correctness*: We want unique container IDs, set expresses this intent better

4. *Safety*: Defensive programming against potential edge cases in kernel registry

Implementation

Update both locations:

• Line 1391: collect_container_stat()

• Line 1403: collect_process_stat()

Verify that collect_container_stat() and collect_per_container_process_stat() accept both list and set as input (or update type hints if needed).

Related

• Parent Epic: BA-4880 (Sprint 26.3 requests)

JIRA Issue: BA-4910