xuanmiao-31's picture
Add InfraBench public task materials + dataset card
02dd0fd verified
|
Raw
History Blame Contribute Delete
1.69 kB

Reference Solution

Step 1 — Identify the degraded node

nodetool status
# → node1 (10.10.1.2) shows DN (Down/Normal)

Step 2 — Diagnose: cluster NIC is down

ssh node1 ip link show enp6s0f1
# → state DOWN — the cluster-internal NIC is offline

The management NIC (enp1s0f0) is still up (SSH works), but node1 lost its cluster gossip link on enp6s0f1.

Step 3 — Restore the NIC

ssh node1 sudo ip link set enp6s0f1 up

Cassandra gossip will detect the NIC coming back and automatically mark node1 as UN within ~30–60 seconds.

Step 4 — Confirm node rejoined

nodetool status
# → All 3 nodes UN

Step 5 — Repair stale data

While node1 was isolated, writes at QUORUM consistency went to node0 and node2. node1 missed those mutations.

ssh node1 nodetool repair bench
# Wait for: "Repair completed successfully"

Step 6 — Verify consistency

cqlsh 10.10.1.2 -e "CONSISTENCY ONE; SELECT * FROM bench.kv;"
# → Should show 5 rows (ids 1–5, including delta and epsilon written during outage)

Root Cause Summary

Physical NIC failure on node1's cluster interface (enp6s0f1) caused the node to become unreachable on the Cassandra gossip/storage network (10.10.1.x). The node continued to function (OS up, management SSH accessible) but was invisible to the cluster. Writes at QUORUM consistency proceeded with 2/3 replicas (node0 + node2), leaving node1 with a stale copy. Bringing the NIC back up restores gossip, and nodetool repair performs Merkle-tree-based anti-entropy sync to reconcile the divergent data.