Deployment artifacts assume one replica: Helm hard-codes replicas: 1 with no affinity or worker count, compose cannot scale, and all replicas would share one log file #1754

What

• deployments/helm/templates/memmachine-deployment.yaml:2 says "(1 replica)" and :16 hard-codes replicas: 1; neither is a chart value. The env list and the env ConfigMap set no MEMMACHINE_WORKERS, so each pod runs one worker. LOG_FILE (:94) is set, but the code reads LOG_PATH (common/configuration/log_conf.py:96).
• deployments/helm/templates/memmachine-service.yaml:20 is type: NodePort with no sessionAffinity, which MCP over HTTP needs today (see the stateful-session issue).
• Logging goes to /app/data/memmachine.log (deployments/helm/templates/memmachine-configmaps.yaml:24) through a plain FileHandler (log_conf.py:117) on the shared RWX PVC, so replicas would interleave writes in one file.
• docker-compose.yml:102 pins container_name: memmachine-app, which blocks docker compose up --scale.
• docs/ has no page on replicas, workers or scaling, while tools/sizing/README.md assumes many API servers.

Expected

Replica count, worker count and affinity are chart values; logs go to stdout by default; compose documents single-instance use or drops the fixed name; a docs page states which components are safe with several replicas and which are not (count cache, config cache, short-term memory, config API, MCP sessions, SQLite stores).

Notes

Read at d6068cdbf (main), paths under packages/server/src/memmachine_server/. Task rather than bug: nothing is wrong for the single-replica deployment the chart describes.

🤖 Written by Claude Code (Claude Fable 5.1) on behalf of @edwinyyyu.