fix: coordinate deployments when shared cache writers are active

docs/shared-cache.md

@@ -171,6 +171,10 @@ History and decisions are capped at 100,000 rows each, aliases at 200,000. Expir
 history/decision cleanup removes at most 256 rows per pass. Access history has
 30-second resolution. Stop all cache processes before upgrading an existing index
 and restart them on the same version; do not mix old and new writers.
+`scripts/deploy.sh` coordinates restarts when existing containers have shared caching
+enabled. For first activation or an explicit budget/configuration change, run
+`COORDINATED_RESTART=1 bash scripts/deploy.sh`. It stops and verifies every old
+application writer before starting replacements. This briefly interrupts service.
 
 ### Performance and validation
 

scripts/deploy.sh

@@ -13,6 +13,7 @@ IMAGE="${IMAGE:-tdurieux/anonymous_github:v2}"
 REPLICAS="${STREAMER_REPLICAS:-4}"
 TIMEOUT="${HEALTH_TIMEOUT:-90}"
 FORCE="${FORCE:-0}"
+COORDINATED_RESTART="${COORDINATED_RESTART:-0}"
 
 log()  { printf '==> %s\n' "$*"; }
 warn() { printf '!!  %s\n' "$*" >&2; }
@@ -79,6 +80,31 @@ if [[ $legacy_alerts == 1 ]]; then
     exit 0
 fi
 
+# Cache index upgrades and budget changes must not overlap old/new writers.
+# Set this explicitly when enabling shared caching for the first time.
+for app in "${apps[@]}"; do
+    if docker inspect -f '{{range .Config.Env}}{{println .}}{{end}}' "$app" | grep -qx 'SHARED_CACHE_ENABLED=true'; then
+        COORDINATED_RESTART=1
+    fi
+done
+if [[ $COORDINATED_RESTART == 1 ]]; then
+    log "Coordinated application restart for shared cache compatibility"
+    docker compose stop -t 30 anonymous_github streamer
+    for app in "${apps[@]}"; do
+        [[ $(docker inspect -f '{{.State.Running}}' "$app") == false ]] || { warn "cache writer $app is still running"; exit 1; }
+    done
+    docker compose up -d --no-deps --force-recreate --scale "streamer=${REPLICAS}" streamer
+    mapfile -t news < <(list_streamers)
+    [[ ${#news[@]} == "$REPLICAS" ]] || { warn "streamer replica count does not match ${REPLICAS}"; exit 1; }
+    for app in "${news[@]}"; do wait_healthy "$app"; done
+    docker compose up -d --no-deps --force-recreate anonymous_github
+    api="$(docker compose ps -q anonymous_github)"
+    [[ -n $api ]] || { warn "API container missing after coordinated restart"; exit 1; }
+    wait_healthy "$api"
+    docker compose ps
+    exit 0
+fi
+
 # 2. Cold start
 mapfile -t olds < <(list_streamers)
 if [[ ${#olds[@]} -eq 0 ]]; then

test/deploy-owner-alert-migration.test.js

@@ -22,6 +22,7 @@ if (args[0] === "image" && args[1] === "inspect") {
   finish(state.built ? "sha:new" : "sha:old");
 }
 if (args[0] === "inspect") {
+  if (args.some(a => a.includes("Config.Env"))) finish(state.sharedCache ? "SHARED_CACHE_ENABLED=true\\n" : "");
   if (args.some(a => a.includes("Labels"))) finish(state.oldProtocol || "<no value>");
   if (args.includes("{{.State.Running}}")) finish(state.stopped && !state.refuseStop ? "false" : "true");
   finish("healthy");
@@ -43,12 +44,26 @@ finish("unexpected docker call", 2);
       const result = spawnSync("bash", [path.join(directory, "scripts/deploy.sh")], {
         encoding: "utf8", timeout: 8000,
         env: { ...process.env, PATH: directory + path.delimiter + process.env.PATH,
-          FORCE: String(options.force ?? 1), STREAMER_REPLICAS: "4", TASK_DOCKER_STATE: statePath, TASK_DOCKER_LOG: logPath },
+          COORDINATED_RESTART: String(options.coordinated ?? 0), FORCE: String(options.force ?? 1), STREAMER_REPLICAS: "4", TASK_DOCKER_STATE: statePath, TASK_DOCKER_LOG: logPath },
       });
       return { ...result, calls: fs.readFileSync(logPath, "utf8").trim().split("\n").map(line => JSON.parse(line)) };
     } finally { fs.rmSync(directory, { recursive: true, force: true }); }
   }
 
+  for (const sharedCache of [true, false]) it(`coordinates cache rollout with ${sharedCache ? "existing" : "new"} caching`, () => {
+    const result = deploy({ oldProtocol: "recipient-v2", sharedCache, coordinated: sharedCache ? 0 : 1 });
+    expect(result.status, result.stderr).to.equal(0);
+    const stop = result.calls.findIndex(args => args[0] === "compose" && args[1] === "stop");
+    const up = result.calls.findIndex(args => args[0] === "compose" && args[1] === "up");
+    expect(stop).to.be.greaterThan(-1); expect(up).to.be.greaterThan(stop);
+    expect(result.calls[up]).to.include.members(["--force-recreate", "streamer=4"]);
+    expect(result.calls.some(args => args.includes("streamer=8"))).to.equal(false);
+  });
+  it("does not start new cache writers when an old writer fails to stop", () => {
+    const result = deploy({ oldProtocol: "recipient-v2", coordinated: 1, refuseStop: true });
+    expect(result.status).not.to.equal(0);
+    expect(result.calls.some(args => args[0] === "compose" && args[1] === "up")).to.equal(false);
+  });
   it("stops and verifies every legacy writer before starting any migration-capable instance", () => {
     const result = deploy();
     expect(result.status, result.stderr).to.equal(0);