Description

After upgrading from 2025.12.4 to 2026.2.0, the ak worker process memory usage approximately doubled.

Environment

• Previous version: 2025.12.4
• New version: 2026.2.0
• Deployment: Kubernetes (K3s), 2 worker replicas
• Database: PostgreSQL 18.2 via PgBouncer (transaction pooling)
• Redis: 8.6.0

Observed Memory Usage (Prometheus container_memory_working_set_bytes)

The server (ak server) pods did not show a similar increase — they stayed in the 500-800Mi range on both versions.

Impact

We had to increase the worker container memory limit from 1200Mi to 1500Mi. The old 1200Mi limit triggered ContainerMemoryNearLimit alerts at 85% (1023Mi/1200Mi).

Notes

• The jump is too large (~500Mi) to be explained by the Python 3.14 runtime alone
• Possible contributors: new Django apps loaded (WS-Federation, Fleet connector, SSF, lifecycle management, endpoint devices), new chunked_queryset usage, django_postgres_cache backend
• May not be the bug pe se, but I was suprised + monioring caught memory consumption was near limit