saved

docs/diagrams/19_failure_indexer_crash.puml

@@ -0,0 +1,55 @@
+@startuml failure_indexer_crash
+title Indexer Failure → replenish from a Native
+
+participant "Node" as N
+participant "Indexer A (alive)" as IA
+participant "Indexer B (crashed)" as IB
+participant "Native A" as NA
+participant "Native B" as NB
+
+note over N: Active Pool : Indexers = [IA, IB]\\nActive Heartbeat long-lived from IA & IB
+
+== IB Failure ==
+IB ->x N: heartbeat fails (sendHeartbeat err)
+note over N: doTick() dans SendHeartbeat triggers failure\\n→ delete(Indexers[IB])\\n→ delete(IndexerMeta[IB])\\nUnique heartbeat goroutine continue
+
+N -> N: go replenishIndexersFromNative(need=1)
+
+note over N: Reduced Pool to 1 indexers.\\nReplenish triggers with goroutine.
+
+== Replenish from natives ==
+par Fetch pool (timeout 6s)
+    N -> NA: GET /opencloud/native/indexers/1.0\\nGetIndexersRequest{Count: max}
+    NA -> NA: reachableLiveIndexers()\\n(IB absent because of a expired heartbeat)
+    NA --> N: GetIndexersResponse{Indexers:[IA,IC], FillRates:{IA:0.4,IC:0.2}}
+else
+    N -> NB: GET /opencloud/native/indexers/1.0
+    NB --> N: GetIndexersResponse{Indexers:[IA,IC]}
+end par
+
+note over N: Fusion + duplication → candidates = [IA, IC]\\n(IA already in pool → IC new candidate)
+
+par Consensus Phase 1 (timeout 4s)
+    N -> NA: /opencloud/native/consensus/1.0\\nConsensusRequest{Candidates:[IA,IC]}
+    NA --> N: ConsensusResponse{Trusted:[IA,IC]}
+else
+    N -> NB: /opencloud/native/consensus/1.0
+    NB --> N: ConsensusResponse{Trusted:[IA,IC]}
+end par
+
+note over N: IC → 2/2 votes → admit\\nadmittedAt = time.Now()
+
+par Phase 2 — liveness vote (if stable voters )
+    N -> IA: /opencloud/indexer/consensus/1.0\\nIndexerConsensusRequest{Candidates:[IC]}
+    IA -> IA: StreamRecords[ProtocolHB][IC]\\nLastSeen ≤ 120s && LastScore ≥ 30
+    IA --> N: IndexerConsensusResponse{Alive:[IC]}
+end par
+
+note over N: IC confirmed alive → add to pool
+
+N -> N: replaceStaticIndexers(pool={IA,IC})
+N -> IC: SendHeartbeat /opencloud/heartbeat/1.0 (goroutine long-live)
+
+note over N: Pool restaured to 2 indexers.
+
+@enduml