Ceph » rgw

was checking on what AWS does, and could not find any official doc listing,
Just some reference where it says it retries 3 times.

Hi Yuval,
While we still get consensus around retry options/timeout options, we have quorum on at least to change the asyn notification to use cls_Fifo.
So in today's meeting, I proposed that bloomberg can start work on fifo implementation while the other options are being tracked.
Also there was a suggestion to have a fix size for FIFO as well, but there was no value was decided. Again while we decide on the value for FIFO max_size we can start our work on implementing FIFO.
Just putting out here, since you were not there in the meeting. Let me know if any comments.

these are great news!
implementing the queue via FIFO would allow for much larger queue, and may also improve on performance as the FIFO usually utilize 2 OSDs - one for read and one for write.

regarding the question of size. allowing for unlimited FIFO would simplify the implementation of the "2 phase commit (=2pc) queue", as it will not be necessary to reserve and then commit. there will be only "push" to the FIFO, that would fail only if we ran out of space in the pool.
however, it would probably be straight forward to reuse the code of the 2pc_queue in the case of the FIFO.

I had a discussion with Yuval, we decided on the following:
we will implement a limitation for the retries of persistent topic by:
1. adding a maximum retries for the notification
2. adding an expiry by seconds (retry for max 100 seconds)
3. we want to add retry sleep to control the frequency of the retries (retry every 10 seconds)

we will add these configuration globally and per topic

Thanks Ali for getting things moving, we were going to bring this up as we (Bloomberg) have started working on this tracker.
So adding both retry and timeout seems meaningful. to configure this will it be part of CreateTopic or rgw config (global value )
Would be nice know so that we can work along

You are welcome!!
We will do both, one global configured by RGW admin (used in case topic doesn't specify a retry configuration) and another per topic which should override the global configuration.
This is the plan as of now, we are still working on this (and it got delayed a bit because I worked on bug recently that had to be addressed urgently)

Ali Masarwa wrote:

You are welcome!!
We will do both, one global configured by RGW admin (used in case topic doesn't specify a retry configuration) and another per topic which should override the global configuration.
This is the plan as of now, we are still working on this (and it got delayed a bit because I worked on bug recently that had to be addressed urgently)

hey Ali, some confusion here when you say "we are still working on this", does it mean that you'll are working on finalizing the requirements ? or doing the actual development work ? As per initial meetings, we had volunteered to do dev work for this feature.