Concurrent modification error on the same bucket (@v25.3.2)

[tolgaulas]

When on load, the arcadedb http api occasionally response with 503

{"error":"Cannot execute command","detail":"Concurrent modification on page PageId(pigeon/82/22) in file \u0027message_0.82.65536.v0.bucket\u0027 (current v.201 \u003c\u003e database v.202). Please retry the operation (threadId\u003d28)","exception":"com.arcadedb.exception.ConcurrentModificationException"}

I guessing this may be due the backplane engine "nature", the both insert requests ended in the same page (bad luck)..

Questions :

• Can Insert have a retry mechanism (as in batch operations)? (feature request, perhaps)
• Can this have a specific "error" message so that client can differentiate the kind/cause of the error and retry. (which looks like the only way resolvable)
• am I totally mistaken, it's my fault or bug.

[gramian]

I am guessing here, but maybe the type being inserted to needs more buckets?

[tolgaulas]

I am guessing here, but maybe the type being inserted to needs more buckets?

So true, but the server is on a mere 1cpu box with the minimal mem. Yes bucket will help, may save my case. but pls consider.

[tolgaulas]

...exprerience: Adding bucket with Alter type .... +type2 caused problem on live server, root user that added became not-permitted (based on error messages)

Had to delete and recreate index on the type (with another admin user). and a server restart; fixed it; though don't know which worked.

I also noticed, the adding bucket did not create the related in_edges and out_edges files in database folder. But they appered only after the recreate index and/or database restart.

Just fyi.

[tolgaulas]

.. unfortunately, adding one more bucket did not solve the issue.. I think one should addjust the bucket strategies according to the server load which I don't think is very "elegant". @lvca whats say you on this?

[lvca]

Having a retry policy on the command could be a good idea:

1. the client sends a command to change some records
2. there is a conflict and the tx must be retried
3. the http handler instead of propagating the error to the client, it will retry X times (configurable). This behavior is only for the NeedRetryException (like ConcurrentModificationException)

@tolgaulas WDYT?

[tolgaulas]

@tolgaulas WDYT?

Lovely! :)

[lvca]

I've just added a new parameter "retry" (integer) in /command payload. If present will auto retry X times the implicit transaction on the server side. @tolgaulas any chances you can try the latest snapshot with the retry parameter if solves the issues?