Session takeover rejects reconnection with DUPLICATE_SESSION_CONNECTION despite incremented sessionVerId

[pedrosakuma]

Problem

When trading-host restarts and attempts to reconnect with an incremented sessionVerId, matching-platform rejects with DUPLICATE_SESSION_CONNECTION instead of allowing session takeover.

Expected Behavior

Per FIXP spec, when a client reconnects with a higher sessionVerId than the previous connection:

1. TryForceTakeOver() should detect verId > lastRecordedVerId
2. Old session should be evicted
3. New session should be established
4. Orders should remain valid on the exchange

Actual Behavior

• Trading-host persists SessionVerId: 2 in snapshot.json
• On restart, trading-host should send SessionVerId: 3
• Matching-platform rejects with DUPLICATE_SESSION_CONNECTION
• TryForceTakeOver() never triggers (no "taking over sessionId=X" log)
• Session takeover code from PR fix(#488): session takeover on reconnect + responsive heartbeat defaults #491 is not being reached

Evidence

# Trading-host snapshot shows verId=2
cat /var/lib/b3trading/entrypoint-state/FIRM01/snapshot.json
{"SessionVerId":2,...}

# Matching-platform rejects reconnection
Negotiate failed: DuplicateConnection
FixpGateway disconnected, will retry in 5000ms

No "taking over sessionId" message appears in matching logs.

Impact

• Trading-host cannot reconnect to matching after restart (unless matching also restarts)
• Orders appear Working in trading-host API but may not be in exchange book
• Breaks session resilience for production scenarios

Root Cause Analysis Needed

1. Is trading-host actually sending incremented sessionVerId? (Add logging)
2. Is TryForceTakeOver() being called at all?
3. What verId comparison is happening in SessionClaimRegistry?
4. Is PR fix(#488): session takeover on reconnect + responsive heartbeat defaults #491 code actually in the running image?

Files to Investigate

• src/B3.Exchange.Gateway/SessionClaimRegistry.cs - TryForceTakeOver logic
• src/B3.Exchange.Gateway/FixpSession.Negotiate.cs - Session takeover trigger
• src/B3.Exchange.Gateway/FixpSessionManager.cs - Session lifecycle

Reproduction

# Clean start
docker compose down -v && docker compose up -d

# Send order, verify in book
# ... via frontend ...

# Restart only trading-host
docker compose restart trading-host

# Observe: DUPLICATE_SESSION_CONNECTION errors in trading-host logs
# Observe: No "taking over" in matching-platform logs

Related

• PR fix(#488): session takeover on reconnect + responsive heartbeat defaults #491 - Session takeover implementation (merged but not working)
• Issue GTD: reject past-dated ExpireDate at entry time (close clockless post-sweep window) #504 / PR docs: mark GAP-23 (GTD) as done #505 - Fixed STALE marking on session roll

[pedrosakuma]

Fechado via PR #493.

O código do PR #491 estava correto — o path TryForceTakeOver funciona. A issue era de log-level (LogInformation filtrado) ou imagem Docker desatualizada no ambiente do reporter.

O PR #493 adiciona teste E2E + corrige dois bugs de produção encontrados durante o review:

1. High: sessão evicted sobrescrevia o snapshot da nova sessão no persister (kind=SessionTakeOver não excluía SaveStateSnapshotSafe)
2. Medium: falha de persistência no takeover não restaurava o claim da sessão antiga, deixando-a unclaimed