init

2026-10-27 14:29:43 +03:00
commit 1769946c99
24 changed files with 3762 additions and 0 deletions
@@ -0,0 +1,90 @@
+# Default migration target: cnpg | selectel
+# Override per run: TARGET=selectel ./migrate-one.sh …
+# Or per list line: yc-loop-1  loopxxx  selectel
+TARGET="${TARGET:-cnpg}"
+
+# CNPG target (pooler NodePort; use a node that runs a pooler pod)
+CNPG_HOST="${CNPG_HOST:-95.143.178.247}"
+CNPG_PORT="${CNPG_PORT:-31574}"
+CNPG_SSLMODE="${CNPG_SSLMODE:-disable}"
+# 1 = always use CNPG_HOST (public VIP); 0 = try pooler pod hostIP first
+CNPG_PREFER_CONFIG_HOST="${CNPG_PREFER_CONFIG_HOST:-0}"
+CNPG_POOLER_RW="${CNPG_POOLER_RW:-pooler-cnpg-wilix-rw}"
+
+# Admin role for CREATE ROLE / CREATE DATABASE (kubectl --context local)
+CNPG_KUBE_CONTEXT="${CNPG_KUBE_CONTEXT:-local}"
+CNPG_NAMESPACE="${CNPG_NAMESPACE:-cnpg-loop-wilix}"
+CNPG_CLUSTER="${CNPG_CLUSTER:-loop-cnpg-wilix-1}"
+CNPG_ADMIN_SECRET="${CNPG_ADMIN_SECRET:-cnpg-vizor-admin-role-secret}"
+
+# Selectel Managed PostgreSQL (create user/pass/db in UI = same as MM pod secret)
+SELECTEL_HOST="${SELECTEL_HOST:-}"
+SELECTEL_PORT="${SELECTEL_PORT:-5432}"
+SELECTEL_SSLMODE="${SELECTEL_SSLMODE:-require}"
+# Source (YC MDB) SSL for pg_dump inside transfer Job
+SRC_SSLMODE="${SRC_SSLMODE:-disable}"
+
+# Source Mattermost installs
+SOURCE_NS="${SOURCE_NS:-loop}"
+JOB_NS="${JOB_NS:-loop}"
+# Pin minor to avoid pg_dump/psql \\restrict mismatches across floating :15 tags
+PG_IMAGE="${PG_IMAGE:-postgres:15.8}"
+JOB_TTL_SECONDS="${JOB_TTL_SECONDS:-86400}"
+# pgxfer Job resources (cluster often tight on memory)
+XFER_CPU_REQUEST="${XFER_CPU_REQUEST:-200m}"
+XFER_CPU_LIMIT="${XFER_CPU_LIMIT:-2}"
+XFER_MEMORY_REQUEST="${XFER_MEMORY_REQUEST:-512Mi}"
+XFER_MEMORY_LIMIT="${XFER_MEMORY_LIMIT:-2Gi}"
+
+# migrate-list parallelism
+PARALLELISM="${PARALLELISM:-1}"
+
+# migrate-one local wait for Job (seconds). 0 = wait forever (manual monitoring).
+MIGRATE_JOB_TIMEOUT="${MIGRATE_JOB_TIMEOUT:-86400}"
+# On wait timeout while Job still Running: keep Job (do not delete / mark failed)
+MIGRATE_KEEP_JOB_ON_TIMEOUT="${MIGRATE_KEEP_JOB_ON_TIMEOUT:-1}"
+
+# Inside transfer Job: retry full dump|restore after short network blips (still restarts copy).
+TRANSFER_MAX_ATTEMPTS="${TRANSFER_MAX_ATTEMPTS:-30}"
+TRANSFER_RETRY_SLEEP_SEC="${TRANSFER_RETRY_SLEEP_SEC:-60}"
+# libpq TCP keepalives (seconds) — reduces silent idle drops on long COPY/INDEX
+PG_KEEPALIVES_IDLE="${PG_KEEPALIVES_IDLE:-30}"
+PG_KEEPALIVES_INTERVAL="${PG_KEEPALIVES_INTERVAL:-10}"
+PG_KEEPALIVES_COUNT="${PG_KEEPALIVES_COUNT:-9}"
+
+# Transfer mode:
+#   stream — pg_dump | psql (default; retry = full re-dump from MDB)
+#   file   — pg_dump | tee >(zstd>PVC) | psql; on blip re-restore from file (no MDB re-dump)
+TRANSFER_MODE="${TRANSFER_MODE:-stream}"
+# file-mode volume: pvc | emptydir (emptydir = node disk; survives blips inside same pod only)
+DUMP_VOLUME="${DUMP_VOLUME:-pvc}"
+DUMP_PVC_SIZE="${DUMP_PVC_SIZE:-80Gi}"
+DUMP_STORAGE_CLASS="${DUMP_STORAGE_CLASS:-yc-network-ssd}"
+DUMP_EMPTYDIR_SIZE_LIMIT="${DUMP_EMPTYDIR_SIZE_LIMIT:-50Gi}"
+# zstd | gzip | none  (postgres:15.8 image already has zstd + gzip)
+DUMP_COMPRESS="${DUMP_COMPRESS:-zstd}"
+DUMP_COMPRESS_LEVEL="${DUMP_COMPRESS_LEVEL:-3}"
+# Keep PVC + dump after success (1) so you can re-restore manually
+DUMP_KEEP_PVC="${DUMP_KEEP_PVC:-1}"
+
+# cutover-one: MM needs long plugin sync after DB switch
+CUTOVER_LIVENESS_INITIAL_DELAY="${CUTOVER_LIVENESS_INITIAL_DELAY:-600}"
+CUTOVER_LIVENESS_FAILURE_THRESHOLD="${CUTOVER_LIVENESS_FAILURE_THRESHOLD:-60}"
+CUTOVER_READINESS_INITIAL_DELAY="${CUTOVER_READINESS_INITIAL_DELAY:-120}"
+CUTOVER_READINESS_FAILURE_THRESHOLD="${CUTOVER_READINESS_FAILURE_THRESHOLD:-60}"
+CUTOVER_MEMORY_LIMIT="${CUTOVER_MEMORY_LIMIT:-4Gi}"
+CUTOVER_MEMORY_REQUEST="${CUTOVER_MEMORY_REQUEST:-512Mi}"
+CUTOVER_CPU_REQUEST="${CUTOVER_CPU_REQUEST:-200m}"
+# wait loop: attempts * sleep (default ~30 min)
+CUTOVER_WAIT_ATTEMPTS="${CUTOVER_WAIT_ATTEMPTS:-180}"
+CUTOVER_WAIT_SLEEP_SEC="${CUTOVER_WAIT_SLEEP_SEC:-10}"
+# kill idle CNPG backends left by crashed MM/session-pooler clients
+CUTOVER_STALE_IDLE_MINUTES="${CUTOVER_STALE_IDLE_MINUTES:-5}"
+
+# Files relative to this directory
+TRANSFER_LIST="${TRANSFER_LIST:-transfer-list.txt}"
+DONE_FILE="${DONE_FILE:-done.txt}"
+FAILED_FILE="${FAILED_FILE:-failed.txt}"
+# Unified migration table (source of truth for status / sizes / batches)
+REGISTRY_FILE="${REGISTRY_FILE:-registry.tsv}"
+LOG_DIR="${LOG_DIR:-logs}"