Problem in Storlet cache / container updates

[Eran Rom]

Updating a storlet via swift PUT to the storlet container do not trigger its update in a multi-cluster env.
Can this be related to time difference between machines?

[Hamdi Roumani (roumani)]

We chatted about this in the storlet IRC meeting (Dec 9, 2015).

We believe the problem has to do with how we decide if the cache needs updating, namely:

- We HEAD the object to see if the user has access and to compare timestamps with the cache.

The problem (I use that word loosely) is we are not using X-newest, so in some scenarios not all versions of the object (e.g. the 3 replicas) would have been updated, implying we'll think nothing has changed.

Even X-newest is no guarentee when handoffs are involved.

But it should reduce the window and perf isn't very important here one would argue, if it is we should re-consider this fix.

I'll probably propose a patch but if someone beets me to it, good on em ; - )

Thanks!

[Eran Rom (eranr)]

Here is a debate I have with myself:
AFAIK, x-newest implies more time as all replicas are being addressed, and we HEAD the storlet object with every storlet execution request.
On the one hand, I wonder if we want an extra header (perhaps even x-newest) from the user to make us do the HEAD with x-newest
On the other, might be that the latency incurred by HEAD + x-newest is negligible compared to storlet invocation