| # Cassandra Dead Node Removal | |
| ## Situation | |
| You are the on-call SRE for a 3-node Cassandra 4.1.x cluster running on bare-metal servers. An alert fired 15 minutes ago: one node has completely stopped responding. Your investigation suggests a hardware-level failure — the node's NIC is down and there is no out-of-band management access available. | |
| **Cluster nodes:** | |
| - `node0` — 10.10.1.1 (you are here) | |
| - `node1` — 10.10.1.2 (unreachable — hardware failure) | |
| - `node2` — 10.10.1.3 (`ssh node2` works) | |
| **Keyspace:** `bench` | |
| **Table:** `bench.kv` (schema: `id int PRIMARY KEY, val text`) | |
| **Replication factor:** 3 | |
| **Consistency level in use:** QUORUM | |
| ## Symptoms | |
| - `nodetool status` shows node1 as `DN` (Down/Normal) | |
| - SSH to node1's cluster IP (10.10.1.2) is unreachable | |
| - The cluster is running degraded; QUORUM reads may be at risk if another node fails | |
| ## Your Task | |
| node1 is **permanently dead** — there is no recovery path. Your job is to restore the cluster to a healthy state so it can safely tolerate future failures. | |
| Specifically: | |
| 1. Confirm the cluster's current state and identify the dead node | |
| 2. Remove the dead node from the cluster ring cleanly | |
| 3. Verify data is fully accessible at QUORUM consistency from both surviving nodes | |
| 4. Ensure data is consistent across the two remaining nodes | |
| ## Available Tools | |
| - `nodetool` — Cassandra ring management | |
| - `cqlsh 10.10.1.1` or `cqlsh 10.10.1.3` — CQL shell | |
| - `ssh node2` — access to the second surviving node | |
| - Standard Linux tools | |
| ## Constraints | |
| - node1 is **permanently lost** — do not attempt to resurrect it | |
| - node0 and node2's Cassandra processes must remain running throughout | |