fix(memory-plugin): gate sidecar spawn on cutover-active

PR #2906 spawned the sidecar unconditionally on every tenant boot. The
plugin's first migration runs \`CREATE EXTENSION vector\` which fails
on tenant Postgres without pgvector preinstalled — every staging
tenant redeploy aborted at the 30s health gate. CP fail-fast kept
running tenants on the prior image (no outage), but the new image
was DOA.

Caught on staging redeploy 2026-05-05 19:23 with
\`pq: extension "vector" is not available\`.

Fix: only spawn the sidecar when the operator has flipped the cutover
flag — \`MEMORY_V2_CUTOVER=true\` OR \`MEMORY_PLUGIN_URL\` is set.

  * Aligns the entrypoint to the same opt-in posture wiring.go already
    uses (it skips building the client when MEMORY_PLUGIN_URL is empty).
  * Until cutover, the sidecar isn't even running — no migration, no
    health gate, no boot-time pgvector dependency.
  * Operators activating cutover already redeploy with the new env
    vars set; that's when the sidecar starts. By definition they've
    verified pgvector is available before flipping.
  * MEMORY_PLUGIN_DISABLE=1 escape hatch preserved; harness fix #2915
    becomes belt-and-suspenders (still respected).

Both Dockerfile and entrypoint-tenant.sh updated. Behavior change for
existing deployments: zero (cutover env vars still unset → sidecar
still inert, but now also not running).

Refs RFC #2728. Hotfix for #2906; supersedes the migration-path
fragility class (the sidecar isn't doing migrations on tenants that
won't use it).
This commit is contained in:
Hongming Wang 2026-05-05 12:39:03 -07:00
parent 8254bedf30
commit 412dec0d87
2 changed files with 33 additions and 15 deletions

View File

@ -63,21 +63,30 @@ fi
# Memory v2 sidecar (built-in postgres plugin). Co-located with the
# main server so operators flipping MEMORY_V2_CUTOVER=true don't need
# to provision a separate service. Stays inert at the protocol layer
# until that env var is set — the workspace-server's wiring.go skips
# building the client without MEMORY_PLUGIN_URL, so the running plugin
# is a no-op for traffic.
# to provision a separate service.
#
# Env defaults:
# Spawn-gating: only start the sidecar when the operator has indicated
# they want it — either MEMORY_V2_CUTOVER=true OR MEMORY_PLUGIN_URL set.
# Without that signal, the sidecar adds zero value (the platform's
# wiring.go skips building the client too) but pays a real cost: the
# plugin's first migration runs `CREATE EXTENSION vector`, which fails
# on tenant Postgres without pgvector preinstalled and aborts container
# boot via the 30s health gate. Caught on staging redeploy 2026-05-05.
#
# Env defaults (when sidecar IS spawned):
# MEMORY_PLUGIN_DATABASE_URL = $DATABASE_URL (share existing Postgres;
# plugin's `memory_namespaces` / `memory_records` tables coexist
# with `agent_memories` and the rest of the platform schema —
# no conflicts. Operator can override with a separate URL.)
# MEMORY_PLUGIN_LISTEN_ADDR = :9100
# MEMORY_PLUGIN_LISTEN_ADDR = 127.0.0.1:9100
#
# Set MEMORY_PLUGIN_DISABLE=1 to skip launching the sidecar entirely
# (e.g. an operator running the plugin externally on a separate host).
if [ -z "$MEMORY_PLUGIN_DISABLE" ] && [ -n "$DATABASE_URL" ]; then
# Set MEMORY_PLUGIN_DISABLE=1 to force-skip the sidecar even with
# cutover env set (e.g. running the plugin externally on a separate host).
memory_plugin_wanted=""
if [ "$MEMORY_V2_CUTOVER" = "true" ] || [ -n "$MEMORY_PLUGIN_URL" ]; then
memory_plugin_wanted=1
fi
if [ -z "$MEMORY_PLUGIN_DISABLE" ] && [ -n "$memory_plugin_wanted" ] && [ -n "$DATABASE_URL" ]; then
: "${MEMORY_PLUGIN_DATABASE_URL:=$DATABASE_URL}"
: "${MEMORY_PLUGIN_LISTEN_ADDR:=:9100}"
export MEMORY_PLUGIN_DATABASE_URL MEMORY_PLUGIN_LISTEN_ADDR

View File

@ -21,14 +21,23 @@ PORT=3000 HOSTNAME=0.0.0.0 node server.js &
CANVAS_PID=$!
# Memory v2 sidecar (built-in postgres plugin). See Dockerfile entrypoint
# comment for rationale. Stays inert at the protocol layer until the
# operator sets MEMORY_V2_CUTOVER=true; running it is cheap.
# comment for rationale.
#
# Defaults the plugin's DATABASE_URL to the tenant's DATABASE_URL so
# operators don't need to configure two of them. Plugin tables coexist
# with the platform schema.
# Spawn-gating: only start the sidecar when the operator has indicated
# they want it (MEMORY_V2_CUTOVER=true OR MEMORY_PLUGIN_URL set).
# Without that signal, the sidecar adds zero value and risks aborting
# tenant boot via the 30s health gate when the tenant Postgres lacks
# pgvector. Caught on staging redeploy 2026-05-05:
# pq: extension "vector" is not available
#
# Defaults (when sidecar IS spawned): MEMORY_PLUGIN_DATABASE_URL
# falls back to the tenant's DATABASE_URL.
MEMORY_PLUGIN_PID=""
if [ -z "$MEMORY_PLUGIN_DISABLE" ] && [ -n "$DATABASE_URL" ]; then
memory_plugin_wanted=""
if [ "$MEMORY_V2_CUTOVER" = "true" ] || [ -n "$MEMORY_PLUGIN_URL" ]; then
memory_plugin_wanted=1
fi
if [ -z "$MEMORY_PLUGIN_DISABLE" ] && [ -n "$memory_plugin_wanted" ] && [ -n "$DATABASE_URL" ]; then
: "${MEMORY_PLUGIN_DATABASE_URL:=$DATABASE_URL}"
: "${MEMORY_PLUGIN_LISTEN_ADDR:=:9100}"
export MEMORY_PLUGIN_DATABASE_URL MEMORY_PLUGIN_LISTEN_ADDR