Hey @0xMostafa, is it possible that meilisearch is crashing and restarting automatically?
That would make it try to process the same batch repeatedly; this can typically happen when you don't have enough RAM to run the indexing process you're trying to do. You can reduce the number of processed tasks in a single batch by using this experimental parameter.

I have also encountered the same problem. After exiting abnormally, asynchronous tasks are still being processed and cannot be completed. Shouldn't they be processed in batches to control the number of concurrent executions

Hello here

As @irevoire asked, do you see Meilisearch crashing and restarting?

Do you have the same issue with v1.9.0? Also, do you have the same issue without vectorStore if you tried?

Do you have a reproducible example, or an access to machine to share (in private) to help us debug this?

when generating documents in large quantities，meilisearch exit and restart
VectorStore is default in v1.9.0 run at docker
can provide machines or files

when generating documents in large quantities，meilisearch exit and restart

What's the resources of your machines? (RAM CPU).
It's probably a lack of RAM.
You can also try with this option: --max-indexing-memory 0

Let me know if it helps

4 cores 8g、Will restarting the task not continue if it exits abnormally during task execution? Will a restart that has already been queued become in progress and still not execute?

Will restarting the task not continue if it exits abnormally during task execution? Will a restart that has already been queued become in progress and still not execute?

I'm not sure I get your question.
But when restarting, Meilisearch will try to process the last processing task again and again. So can be stuck if there is a lack of resources.

Did you try to increase the RAM? Did you try with --max-indexing-memory 0?

There are many unexecuted tasks piled up, which should be due to insufficient memory or other reasons. The exported program exited, but after restarting, it should continue with the unexecuted tasks, right

As you said, the last task was executed, but the nearly a thousand tasks piled up earlier did not move and remained stuck. Can we improve the number of task executions to control memory and concurrency, and can we restore execution of failed or abnormal tasks

but the nearly a thousand tasks piled up earlier did not move and remained stuck

What is stuck? What are their status? Use this route: https://www.meilisearch.com/docs/reference/api/tasks#get-tasks

If Meiliseach crashes and restart, it will try to re-process the current. But if the tasks is too big for the machine resources, it will crash in loop again and again.

Did you try to increase the RAM? Did you try with --max-indexing-memory 0?

How many documents do you have? How many do you try to index in the stuck task?

--max-indexing-memory 0 Parameters have been added、Data of approximately 100000 * 1300 tasks、finish 300 、other processing、The task queue should not be executed together with quantity control, otherwise it will never finish and crash

Can you try to reduce the number of tasks process at the same time:
https://github.com/orgs/meilisearch/discussions/713

with --experimental-max-number-of-batched-tasks 1 -> to process tasks one by one

Thank you, I'm looking forward to the official version of the parameter to be added to https://www.meilisearch.com/docs/learn/self_hosted/configure_meilisearch_at_launch This is a very useful tool compared to controlling memory --max-indexing-memory.

Yes sorry, I should have suggested it first! I did not realized the batch queue was too big!

I opened an issue here: meilisearch/documentation#2958

I close this issue since it seems to be fixed! Let me know if it's not