Improve startup time of motr applications (MIO in particular)

[u-u-h]

The MIO startup (https://github.com/Seagate/cortx-mio mio_init()) is very slow, e.g., on the sage system. 40s are not unusual.

For user code connecting to MIO this is unacceptable, both because the application may have a shorter runtime than that altogether, and because the MIO init is blocking (rightly so), so startup delay accumulates with other initializations that may be done. (Note that I'm not suggesting to make it non-blocking; this would only rarely solve the issue, and complicate user code which then cannot rely on MIO being ready.)

[u-u-h]

^^ @nezzzu maybe?

[nezzzu]

@truptiatseagate @andriytk This is the same clovis_init() or m0client_init() that is causing the delay in mio_init(). What is the current status of the fix? When can we expect the fix to be available in the main branch?

[andriytk]

@nezzzu see here - Seagate/cortx-hare#1737.

[hessio]

Is it ok to close this issue now based on this PR Seagate/cortx-hare#1780

[andriytk]

No, not yet. The mentioned PR improves the situation - yes, but it does not fix the problem entirely.

Also, we had to decrease the number of client slots per client node in the Sage cluster from 16 to 4 to further decrease the startup delay, which is 4 secs now. But it should be less than 1 sec.

[hessio]

ok thanks for the update

[chandradharraval]

@u-u-h , @andriytk - Is this still valid issue?

[u-u-h]

I can't test with current versions since the cluster we used has been dismantled. @nezzzu may be able to comment on this, he could reproduce it, and may have access to a current motr installation.

[chandradharraval]

Hi @nezzzu,
Is this still reproducible?

[cortx-admin]

ganesan umanesan commented in Jira Server:

[~530903] The Sage cluster is in the process of being set up as part of the IOSEA project. It is estimated to be back online by October end. I will verify the startup time again once the cluster becomes available.

[shailesh-vaidya]

Closing as an obsolete