File size: 3,143 Bytes
02dd0fd
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
# Reference Solution

## Step 1 — Assess the cluster state

```bash

nodetool status

# → Status/State shows:

# UN  10.10.1.1   ...  <uuid-node0>

# DN  10.10.1.2   ...  <uuid-node1>   ← dead node

# UN  10.10.1.3   ...  <uuid-node2>

```

Note the UUID in the last column of the `DN` row — this is node1's **Host ID**.

## Step 2 — Confirm node1 is unreachable

```bash

ssh -o ConnectTimeout=10 node1

# → ssh: connect to host node1 port 22: Connection timed out

# Confirmed: no SSH access, no recovery possible

```

## Step 3 — Remove the dead node from the ring

```bash

# Extract the Host ID automatically

HOST_ID=$(nodetool status | grep '^DN' | awk '{print $(NF-1)}')

echo "Removing host: $HOST_ID"



nodetool removenode "$HOST_ID"

# Output: "RemovalStatus=COMPLETED" (may take 30–120s while data streams)

```

`removenode` evicts the dead node's token ranges and streams any data it exclusively held to the surviving nodes. Unlike `decommission`, it does not require the node to be running.

## Step 4 — Verify ring is clean

```bash

nodetool status

# → Exactly 2 UN nodes, 0 DN nodes:

# UN  10.10.1.1   ...

# UN  10.10.1.3   ...

```

## Step 5 — Verify data integrity at QUORUM

```bash

cqlsh 10.10.1.1 -e "CONSISTENCY QUORUM; SELECT COUNT(*) FROM bench.kv;"

# → 5



cqlsh 10.10.1.3 -e "CONSISTENCY QUORUM; SELECT COUNT(*) FROM bench.kv;"

# → 5

```

QUORUM with RF=3 requires `ceil(3/2) = 2` responses. Both surviving nodes hold complete replicas after `removenode` streaming, so QUORUM reads are satisfied.

## Step 6 — Repair (best practice)

```bash

nodetool repair bench

# Wait for: "Repair command #N finished" in /var/log/cassandra/debug.log

```

After `removenode`, repair runs Merkle-tree anti-entropy between the two surviving nodes to eliminate any remaining inconsistencies.

## Root Cause Summary

node1 suffered a complete hardware failure — both its cluster NIC and OS-level SSH daemon became unreachable. This is distinct from the NIC split-brain scenario: there is no OS running to accept `ip link set up`, no IPMI BMC accessible to power the node back on. The only correct response is to treat the node as permanently dead and remove it from the Cassandra ring via `nodetool removenode`.

## Key Concepts

| Operation | When to use | Node state required |
|-----------|-------------|---------------------|
| `nodetool decommission` | Planned removal (node is healthy) | Node must be **UP** and reachable |
| `nodetool removenode <HOST_ID>` | Emergency removal (node is dead) | Node must be **DN** or unreachable |

**Why not `decommission`?** It streams data *from* the leaving node, which requires the node to be running and reachable over the cluster network. Since node1's cluster NIC is down and SSH is blocked, `decommission` would hang indefinitely.

**Why QUORUM still works?** With RF=3 and 2 surviving nodes, each row has 2 physical replicas. QUORUM requires `ceil(RF/2) = 2` replicas — exactly what we have. Once `removenode` completes its streaming pass, both survivors are guaranteed to hold every row.