Commit Graph
9 Commits
Author SHA1 Message Date
Naeel d1d1bffd7c v0.1.13: Strategy Recreate + preStop + liveness fix (ERR-SQS-06 root cause)
- ensureDeployment: Strategy Recreate (not RollingUpdate) — prevents H2 file lock
  when two pods mount same PVC simultaneously during rollout restart
- preStop: sleep 3 — graceful H2 shutdown before SIGTERM
- livenessProbe timeoutSeconds: 3 — prevents false positive on GC pause
- terminationGracePeriodSeconds: 15
- doc: thinking log, ERR-SQS-06, progress.md, architecture SVG schema
2026-04-09 07:49:11 +03:00
Naeel 3adc0d8323 sqs-operator v0.1.12: ingress /queues/* -> elasticmq-ui (Next.js basePath fixes) 2026-04-09 06:17:21 +03:00
Naeel a04d72052e sqs-operator v0.1.11: fix ensureService UI port, FILE_LOCK=NO for H2 2026-04-08 22:17:41 +03:00
Naeel 7ea7e9b360 sqs-operator v0.1.9: fix SQS_ENDPOINT context-path, add sidecar to source
- ensureDeployment: sidecar elasticmq-ui добавлен в Go-код (не Python патч)
- SQS_ENDPOINT=http://localhost:9324/sqs/{tenantID} — с context-path
- ElasticMQ не отвечает на root /, только на /sqs/{tenantID}
- Sidecar условный: добавляется только при spec.enableUI=true
2026-04-08 20:08:25 +03:00
Naeel 657fc33119 sqs-operator v0.1.8: fix UI assets ingress (/_next/ path-based routing)
- ensureIngressUIAssets: Prefix ingress /_next/ без rewrite -> UI port 3000
- Next.js basePath="" захардкожен — static assets идут без тенантного префикса
- Решение через отдельный ingress объект (один на тенант, контент идентичен)
2026-04-08 19:40:42 +03:00
Naeel 64626659a9 sqs-operator v0.1.7: web UI sidecar, auto-create-queues, fix A06 long polling
- api/v1alpha1: добавлены поля EnableUI, AutoCreateQueues в QueueServiceSpec
- elasticmq_config.go: GenerateConfig принимает autoCreateQueues, добавляет HOCON блок
- controller.go: sidecar elasticmq-ui (port 3000), ensureIngressUI, ensureService UI port
- test_v2_suite.sh: A06 теперь PASS — PurgeQueue перед long polling тестом
- CRD обновлён: make install применён в кластере

Test run v0.1.7: 35 PASS / 3 FAIL / 4 WARN / 2 SKIP(WONTFIX)
2026-04-08 19:11:26 +03:00
Naeel c132c68d74 sqs-operator v0.1.6: SH02/SH03/SH04 fix + test_v2_suite + v2 test results
- v0.1.5: ensureHealthy checks all 4 resources (Deployment/Service/ConfigMap/Ingress)
  Service/ConfigMap/Ingress now auto-recreate in 2-4s when manually deleted
- v0.1.6: revert MT03 configuration-snippet (nginx blocks risky annotations by default)
  Tenant isolation deferred to Keycloak JWT in production
- test_v2_suite.sh: 8 phases, 52 tests, timing/resources/concurrent/30min marathon
- Results: 40 PASS / 4 FAIL / 7 WARN (known: OOM restart under burst load)
  SH recovery: Deployment=48s Service=2s ConfigMap=4s Ingress=2s All3=2s
  Marathon: 11815 iter, 10830 sent, 2 pod restarts (OOM under load)
2026-04-07 19:14:00 +03:00
Naeel 336ee7b869 fix(sqs-operator): persistence, OOM, Ingress routing (v0.1.2→v0.1.4)
Problem 1: elasticmq-native does not support H2 JDBC persistence
- GraalVM native build excludes H2 driver
- Fix: switch to softwaremill/elasticmq:1.7.1 (JVM image)

Problem 2: OOMKilled — JVM requires >150MB, spec.MemoryMB=64 too low
- Fix: enforce minimum 256Mi for JVM; add -Xmx (75% of limit)

Problem 3: AccessDeniedException on /data — PVC mounted as root:root
- JVM image runs as uid=999 (elasticmq)
- Fix: add fsGroup=999 to PodSecurityContext

Problem 4: 404 through HTTPS — nginx rewrite stripped context-path
- ElasticMQ JVM listens at context-path (/sqs/{tenantId})
- Native image tolerated /, JVM does not
- Fix: remove rewrite-target annotation, use plain Prefix path

Result: S6 PASS — 8 queues + messages survived graceful pod kill
2026-04-07 15:34:15 +03:00
Naeel 66dcd99465 refactor(sqs-operator): переделка через Operator SDK v1.37.0
- operator-sdk init + create api (QueueService kind, group sqs, v1alpha1)
- Стандартная kubebuilder структура: cmd/, internal/controller/, config/
- CRD types с kubebuilder маркерами (validation, defaults, printcolumns)
- Reconciler перенесён из ручного кода в internal/controller/
- controller-gen v0.17.0 (совместимость с Go 1.26.1)
- Автогенерация: deepcopy, CRD YAML, RBAC ClusterRole
- Удалены все ручные файлы (controllers/, main.go, deployments/)
2026-04-07 14:25:14 +03:00