776767 - differences between funkload and grinder tests

Status: VERIFIED FIXED

(Cloud Services Graveyard :: Server: Sync, defect)

Description

[Ryan Kelly [:rfkelly]]

I have done a comparison of the funkload and grinder loadtest scripts, and there are some differences worth noting/thinking about:

   1) grinder POSTs batches of 10 items; funkload POSTs batches of 100 items.

   2) grinder doesn't hit the "tabs" collection; funkload does hit it.

   3) grinder occasionally does a DELETE; funkload never does a DELETE.

   4) grinder will sometimes POST an update to an existing record; funkload always creates new records.

   5) grinder often requests /meta/global; funkload never will.


Item (1) is probably enough to explain the failing loadtests we have been seeing, as this produces a ~10 times increase in write operations in the database.  Discussion suggests that 100 items per batch is closer to realistic behaviour, so I will file a separate bug to hash that out.

We should think about the other items and whether we need to port the grinder behaviour over to funkload.  At least item (3) should definitely be ported over.

Comment 1

[Ryan Kelly [:rfkelly]]

Most of these are included in the patch to Bug 782002

Comment 2

[Ryan Kelly [:rfkelly]]

Closing this out, the majority of these were fixed by Bug 782002 and we have more specific bugs open for the remaining actions.

Comment 3

[James Bonacci [:jbonacci]]

:rfkelly can you mark this as Resolved/Fixed then?

Comment 4

[Ryan Kelly [:rfkelly]]

Whoops, must have added the comment but forgot to update the status

Comment 5

[James Bonacci [:jbonacci]]

Thanks. Based on Comment 2, marking this as Verified.