[CASSANDRA-14981] Cassandra fires NoHostAvailable exception, but trigger receives update

• Type: Bug
• Status: Open
• Priority: Normal
• Resolution: Unresolved

Environment

cassandra 3.11.3

• Severity: Normal
• Assignee: Unassigned
• Reporter: Timur Tibeyev
• Created: 15/Jan/19 04:59
• Updated: 16/Apr/19 09:29

Description

Cluster

I have cluster with three nodes, cluster is created via `docker-compose`.

All nodes in the cluster have `UN` (up normal)  status, checked with `nodetool status`.

 

Trigger

Trigger receives event and produces `Kafka` record.

All three nodes have same trigger.

 

Table

Here is my CQL query, `users` table replication factor is 2

Consistency level is one

```

CREATE KEYSPACE mytestdb WITH replication = { 'class':'SimpleStrategy', 'replication_factor':2 };
USE mytestdb;

CREATE TABLE IF NOT EXISTS users (
id uuid,
username text,
phone text,
lastname text,
firstname text,
PRIMARY KEY (id)
);

CREATE TRIGGER kafka_trigger ON users USING 'org.company.CassandraTrigger';

```

 

Problem

Everything worked fine so I wanted to check edge cases, I intentionally shut down 2 nodes, to check how `Cassandra` and trigger will perform.

I started to make queries to single working node.

Some inserts were processed correctly, but for some of them `NoHostAvailable` exception occurred, but trigger received this "failed" insert event and produced normal `Kafka` record

The problem is - data was not saved to `Cassandra`, but trigger received update.