[LU-940] Many places in the test framework issue 3 syncs [sync; sync; sync] do flush the filesystem before [for example] failover.

• Status: Resolved
• Project: Lustre
• Type: Bug
• Priority: Major
• Reporter: Chris Gearing (Inactive)
• Assignee: WC Triage
• Resolution: Low Priority
• Severity: 3

Description

Many places in the test framework issue 3 syncs [sync; sync; sync] do flush the filesystem before [for example] failover.

This 3 sync strategy should not be required, is not what a user would do and so is either unneccessary or hides unlying sync issues.

The multiple calls to sync should be remove.

Comments

[Mikhail Pershin]

I thought about the same when saw syncs before failover are being done implicitly. This is bad practice because it makes our tests working fine, but nobody will do sync on customer clusters before reboot, so we are getting green test results by moving problems on customer side which is not acceptable. Our green tests make just no sense if bugs are hidden.

Meanwhile there is 3 sync in replay-barrier which is artificial construction just to make sure all data is on disk before we continue. It is used mostly in our unit tests to check various recovery cases with strong operation ordering and state, so it is required to have syncs there otherwise operation state is undefined - committed or not.

[Swapnil Pimpale (Inactive)]

There were a couple of places in test framework which used 3 syncs in a row.
I have fixed those in this patch (http://review.whamcloud.com/#/c/7280)

[Swapnil Pimpale (Inactive)]

I somehow missed Mikhail's comment above.
Since the 3 sync strategy in replay_barrier is an artificial construction, this patch is not required.
Abandoned it.