Elasticache Redis with cluster mode disabled but failover enabled

[cbazureau]

We are facing a READONLY You can't write against a read only replica error (followed by a Socket closed unexpectedly error) when we want to write on our AWS ElastiCache Redis cluster (Cluster mode off) when multi-AZ and failover are enabled.

On redis server side :

On client (nodejs) side :

2 persistant connexions, one for the "master" (called Primary Endpoint on AWS), one for the "replica" (called Reader Endpoint on AWS)

const ACTIVATE_READ = true;

const writeClient = redis.createClient({ url: '<master-endpoint>', ... });
const readClient = ACTIVATE_READ ? redis.createClient({ url: '<master-endpoint>', ... }) : writeClient;

await writeClient.connect();
if (ACTIVATE_READ) await readClient.connect();

await readClient.get('foo');
await writeClient.set('foo', 'bar', { EX: 30 }); // Fails here if ACTIVATE_READ is true
await readClient.get('foo');
await writeClient.quit();
if (ACTIVATE_READ) await readClient.quit();

• No issue with the "replica" connexion when multi-AZ and failover are enabled or disabled
• No issue with the "master" connexion when multi-AZ and failover are disabled or when "master" connexion is used as the only connexion (ACTIVATE_READ = false)
• When multi-AZ and failover are enable, node-redis seems to go to the wrong client (using the readClient instead of the writeClient).

Questions :

• Does node-redis support the AWS failover mode, or did i miss something on the configuration ?
• In case of failover, is there an internal mechanism in node-redis that ensure to switch to the good node.

Environment:

• Node.js Version: 16 and 18 (at least)
• Redis Server Version: AWS ElastiCache Redis 6.2.4 with Multi-AZ + failover enable (2 nodes)
• Node Redis Version: 4.5.0
• Platform: docker alpine-node

[cbazureau]

FYI i've switched to ioredis with the same configuration (2 connexions, one to write / one to read) and it works

[af-andy]

Anybody else run into this and find a solution?