Stabilise CLI P2P smoke readiness

.github/workflows/cli-p2p-compose-smoke.yml

@@ -60,10 +60,17 @@ jobs:
           BENCH_BIN_SIZE_BYTES: '512'
           BENCH_COUCHDB_RTT_MS: ${{ inputs.couchdb_rtt_ms || '20' }}
           BENCH_SYNC_TIMEOUT: '180'
-          BENCH_PEERS_TIMEOUT: '90'
+          BENCH_PEERS_TIMEOUT: '20'
+          LIVESYNC_P2P_RELAY_READY_TIMEOUT_MS: '60000'
           BENCH_LIVESYNC_TEST_TEE: '0'
         run: docker compose -f test/bench-network/compose.yml run --rm bench-runner
 
+      - name: Show Compose diagnostics
+        if: failure()
+        run: |
+          docker compose -f test/bench-network/compose.yml ps
+          docker compose -f test/bench-network/compose.yml logs --no-color couchdb nostr-relay
+
       - name: Upload benchmark results
         if: always()
         uses: actions/upload-artifact@v4

src/apps/cli/testdeno/bench-p2p.ts

@@ -179,9 +179,9 @@ async function main(): Promise<void> {
         await host.waitUntilContains("P2P host is running", 20000);
         const hostReadyElapsed = nowMs() - hostReadyStart;
 
-        const peerDiscoveryStart = nowMs();
+        const peerDiscoveryCommandStart = nowMs();
         const peer = await discoverPeer(clientVault, clientSettings, config.peersTimeoutSeconds);
-        const peerDiscoveryElapsed = nowMs() - peerDiscoveryStart;
+        const peerDiscoveryCommandElapsed = nowMs() - peerDiscoveryCommandStart;
 
         const syncStart = nowMs();
         await runCliOrFail(
@@ -223,7 +223,10 @@ async function main(): Promise<void> {
             binFileCount: seedFiles.binCount,
             mirrorElapsedMs: Number(mirrorElapsed.toFixed(1)),
             hostReadyElapsedMs: Number(hostReadyElapsed.toFixed(1)),
-            peerDiscoveryElapsedMs: Number(peerDiscoveryElapsed.toFixed(1)),
+            peerDiscoveryTimeoutSeconds: config.peersTimeoutSeconds,
+            peerDiscoveryCommandElapsedMs: Number(peerDiscoveryCommandElapsed.toFixed(1)),
+            peerDiscoveryNote:
+                "p2p-peers waits for the requested timeout before printing discovered peers, so this is command duration, not first-peer latency.",
             syncElapsedMs: Number(syncElapsed.toFixed(1)),
             throughputBytesPerSec: Number((seedFiles.totalBytes / (syncElapsed / 1000)).toFixed(2)),
             throughputMiBPerSec: Number((seedFiles.totalBytes / (syncElapsed / 1000) / 1024 / 1024).toFixed(4)),

test/bench-network/README.md

@@ -68,6 +68,10 @@ The current CouchDB latency model is the existing HTTP proxy inside
 latency, but it does not model packet loss, jitter, MTU, bandwidth limits,
 bufferbloat, or VPN encapsulation.
 
+For P2P runs, `BENCH_PEERS_TIMEOUT` is passed to `p2p-peers`. That command waits
+for the requested observation window before printing discovered peers, so the
+reported peer discovery command time should not be read as first-peer latency.
+
 ## Latency sweep
 
 To run P2P once and CouchDB at several requested RTT values:

test/bench-network/compose.yml

@@ -45,6 +45,11 @@ services:
         exec /app/strfry --config /tmp/strfry.conf relay
     tmpfs:
       - /app/strfry-db:rw,size=256m
+    healthcheck:
+      test: ["CMD-SHELL", "nc -z 127.0.0.1 7777"]
+      interval: 2s
+      timeout: 5s
+      retries: 30
 
   coturn:
     image: coturn/coturn:latest
@@ -64,7 +69,7 @@ services:
       couchdb:
         condition: service_healthy
       nostr-relay:
-        condition: service_started
+        condition: service_healthy
     environment:
       BENCH_COMMAND: ${BENCH_COMMAND:-cases}
       BENCH_CASES: ${BENCH_CASES:-couchdb-baseline,p2p-direct-local}
@@ -88,6 +93,7 @@ services:
       BENCH_TETHERING_VPN_RTT_MS: ${BENCH_TETHERING_VPN_RTT_MS:-120}
       BENCH_SYNC_TIMEOUT: ${BENCH_SYNC_TIMEOUT:-300}
       BENCH_PEERS_TIMEOUT: ${BENCH_PEERS_TIMEOUT:-60}
+      LIVESYNC_P2P_RELAY_READY_TIMEOUT_MS: ${LIVESYNC_P2P_RELAY_READY_TIMEOUT_MS:-60000}
       BENCH_LIVESYNC_TEST_TEE: ${BENCH_LIVESYNC_TEST_TEE:-0}
     volumes:
       - ./bench-results:/workspace/src/apps/cli/testdeno/bench-results