Fix: Add keep-alive options to Redis clients to prevent idle timeouts and socket closing.

[nikitas-novatix]

Problem:

When Flowise is deployed with Redis (especially in Queue Mode or environments like Azure Cache for Redis with idle timeouts), Redis connections can be closed by the server after a period of inactivity (e.g., 10-15 minutes). This results in uncaughtException: Socket closed unexpectedly errors, crashing worker processes or causing failures in other Redis-dependent components.

This issue is tracked in #2186.

Solution:

This PR adds keep-alive mechanisms to the Redis clients used throughout Flowise to ensure connections remain active even during idle periods. The following settings were added to the respective client configurations:

• Core Queue Mode (@redis/client): Added socket: { keepAlive: 60000 } and pingInterval: 60000 to the clients used by RedisEventPublisher and RedisEventSubscriber.
• BullMQ (QueueManager - ioredis): Added keepAlive: 60000 and enableReadyCheck: true to the ioredis connection configuration used by BullMQ.
• Other ioredis Clients: Added keepAlive: 60000 and a default retryStrategy to clients used in CachePool, RateLimiterManager, RedisCache, RedisEmbeddingsCache, and RedisBackedChatMemory nodes.
• Other @redis/client Clients: Added socket: { keepAlive: 60000 } and pingInterval: 60000 to the client used in the Redis Vector Store node.

The keep-alive/ping intervals are set to 60000ms (1 minute) as a robust default, significantly shorter than typical server timeouts, ensuring connections are maintained without excessive background traffic.

Testing:

• Successfully reproduced the "Socket closed unexpectedly" error locally using Queue Mode, SQLite, and a Dockerized Redis instance configured with a short (--timeout 10) idle timeout.
• Verified that applying the keep-alive fixes with a short interval (e.g., 5000ms) prevented the original error under the same test conditions.

Closes:

Fixes #2186

[nikitas-novatix]

@HenryHengZJ thanks for your comments.

I did the following changes:

• Added a REDIS_KEEP_ALIVE env variable, 60000 by default (1min).
• Added a one-liner based on your suggestion, but made it robust to mistakes in the environment variable. So if a person sets something that isn't a number, it still runs (without the KeepAlive mechanism)
• Simplified the code for consistency by removing non-necessary parts (pingInterval, RetryMechanism).

[nikitas-novatix]

Hey @HenryHengZJ

I realized the PingInterval is actually necessary for @redis/client instances, as they need both socket.keepAlive (TCP-level) and pingInterval (Redis protocol-level) to prevent timeouts. ioredis handles this differently with just keepAlive. Updated the PR to use REDIS_KEEP_ALIVE for both settings.

I tested this thoroughly this time, sorry for the inconvenience.

I created a new PR with the patch in #4431