When not to use enableAutoPipelining?

[stavalfi]

1. From the docs, it sounds like there are only advantages. So why ioredis does not internally use it without any user-configurations?

• I have a single node cluster with 1 shard (master). Maybe in my specific use case there are only advatanges for enableAutoPipelining?

2. Also, If I send multi, will the commands inside this multi will be send in a "bigger" multi (incase there are some commands waiting in the queue)?

3. With enableAutoPipelining=true, does ioredis send at most only 1 request at the same time?

4. maybe the answers should be posted in the documentation.

---

Thanks for this super great feature!! Good job!!

[TysonAndre]

From the docs, it sounds like there are only advantages. So why ioredis does not internally use it without any user-configurations?

Not a maintainer, but there were still some open issues, some of which were recently fixed in 4.27.7 (#1391) - maybe they're waiting for more real-world testing or a major release before making it a default

Second, if 100 commands are automatically pipelined, and there's a (EDIT: Networking, not redis) error (EDIT: and you get disconnected before the first response is sent), you don't know how many failed, which may be fine in some applications but undesirable in others (e.g. if there were 100 increment commands, you wouldn't know how many (0-100) of the increments actually succeeded or actually failed - having 100 requests not knowing if they succeeded or failed after a spurious networking error is worse (in some applications) than just 1 request with an unknown result (reconnecting then sending the other 99))

• Autopipelining is generally the best choice for high throughput or bursts of requests to process in my experience, though

2. If I send multi, will the commands inside this multi will be send in a "bigger" multi (incase there are some commands waiting in the queue)?

I think the multi is atomic and can't be added in a pipeline together with other commands from briefly reading the lib/autopipelining code (array of commands that forbid autopipelining). I'm not certain if that's correct.

• I'm likely misreading that, though. Maybe the multi is excluded because they aren't actually requests yet

3. With enableAutoPipelining=true, does ioredis send at most only 1 request at the same time?

No, that describes what happens when it's false. It sends many requests at a time without waiting for a response when it's true

[stavalfi]

Thanks for your time @TysonAndre!

if 100 commands are automatically pipelined, and there's an error, you don't know how many failed

How can it be? I will get the result-status of every one of my requests so I will know if a specific request succeed or not. Do I have any mistake?

With enableAutoPipelining=true, does ioredis send at most only 1 request at the same time?

No, that describes what happens when it's false. It sends many requests at a time without waiting for a response when it's true

I'm confused. it's the point of autopipeline to wait for more commands (while there is an ongoing request) and when there will be a response, collect all the pending requests and then go to redis only once?

It seems like I understand the opposite from your explanantion.

[TysonAndre]

With enableAutoPipelining=true, does ioredis send at most only 1 request at the same time?

Actually, I was confused by the name in my previous response. In both cases it's sending multiple requests at the same time before waiting for a response - i.e. https://en.wikipedia.org/wiki/Protocol_pipelining is enabled in both cases

What enableAutoPipelining is doing is grouping the multiple writes of small packets into a single stream write of a larger amount of data, making more efficient use of the network link and sending fewer total tcp packets https://github.com/luin/ioredis#autopipeautopipelininglining