-
Notifications
You must be signed in to change notification settings - Fork 333
Current Capacity is not working properly #404
Comments
My prover keeps starting generating proofs for proposed blocks for "external" proposers while it is not accepting new assign requests on port 9876 (closed).
|
Today node experiences the same.
Prover log:
Proposer log:
|
time to time client returns
While it is |
I identified culprit: As per log below we can see that capacity ReleaseOneCapacity() was called 2 times:
It is caused by #402 |
Did not fix for me.
Capacity at 1
Instances set to 1
Thanks. |
emmm did u restart ur simple-taiko-node? |
Yes, sir. It appears that ZKEVM_CHAIN_INSTANCES_NUM is respected right after start. After proof is generated, 2 proofs are started. And so on. |
Proof generating is stuck and I have restarted 3 times. It works better for sometime and then it encounters the same error.
I have set instances to 2
Current capacity is 5.
|
Additional fix in #412 should fix issues. |
This is being logged as an error, maybe it should be a warning. |
Set --proposeInterval = 1s and got issue with capacity manager, where Proposer was able to assign prover up to x10 times, before capacityManager started taking out capacity.
|
I have pushed a fix here: this should resolve both yours and @davaymne issues. this introduces a new "tempCapacity" on the server, to solve davaymes issue, and it ties the capacity to a |
I can observe that capacity Manager is partially working, but under certain condition it is still failing to control capacity. Set --proposeInterval = 1s and got issue with capacity manager, where Proposer was able to assign prover 10 times (again), before capacityManager started taking out capacity. Please see proposer logs and screenshot of txs:
|
Please modify log as current version could mislead.
I read |
I will improve the logs. Regarding the capacity being taken, I am wondering if your prover was running and caught up when it was assigned? Right now we have the concept of tempCapacity, which is only taken for I am wondering if the issue is potentially here: // If there is newly generated proofs, we need to submit them as soon as possible.
if len(p.proofGenerationCh) > 0 {
end()
return nil
} which is code we have in the Do you have the prover logs from this same proposer spam opration, that would maybe help me diagnose? |
Capacity manager will be removed for A6: #478 |
What: Prover concurrently is generating proofs for more blocks as it was supposed to and status API returns wrong numner of available capacity.
Config:
From log we can see that there are 5 ongoing proof generating (it was supposed to be max 3).
Log:
docker logs -f simple-taiko-node-taiko_client_prover_relayer-1
Status returns incorrect number of available capacity (it was supposed to be 0).
Status:
$ curl http://172.23.0.7:9876/status
The text was updated successfully, but these errors were encountered: