Howard Chu authored Aug 31, 2020 and Quanah Gibson-Mount committed Aug 31, 2020

Don't assume si_cookieState is always newer

Howard Chu authored Aug 31, 2020 and Quanah Gibson-Mount committed Aug 31, 2020

In a standard Refresh present phase, the provider sends no cookie
since it is only listing the entries that existed as of the time
in the cookie the consumer sent. In this case the consumer only
needs to check entryCSNs against its last sent cookie.

Howard Chu authored Aug 26, 2020 and Quanah Gibson-Mount committed Aug 26, 2020

A bit uglier but more straightforward.

Howard Chu authored Aug 22, 2020 and Quanah Gibson-Mount committed Aug 24, 2020

Just poll for available data, same as Persist mode.
Clarify retry/return states from do_syncrep2

Howard Chu authored Jul 29, 2020 and Quanah Gibson-Mount committed Jul 29, 2020

For delta-sync as well as regular sync

Howard Chu authored Feb 22, 2018 and Quanah Gibson-Mount committed Feb 28, 2018

And try another approach. Always write contextCSN updates, but
don't set dont_replicate for updates we want propagated.

Howard Chu authored Jan 31, 2018 and Quanah Gibson-Mount committed Feb 21, 2018

If syncprov is present, only write contextCSN attribute on
actual state changes, not on per-entry modifications.
Continue to update in-memory cookieState. Saves overhead,
syncprov will eventually checkpoint it into the DB anyway.

Howard Chu authored Oct 13, 2017 and Quanah Gibson-Mount committed Oct 17, 2017

Must release cookieState->cs_mutex before invoking backend.
Add a condvar to serialize calls of updateCookie, so we can
release the mutex and still update sequentially.

Also added tid logging, useful in conjunction with
7ab0e1aff0cc48cdfb299ca7dbd27900a9e3d1a8

Howard Chu authored Oct 05, 2017 and Quanah Gibson-Mount committed Oct 17, 2017

Make sure it's last in callback stack

Ondřej Kuzník authored Apr 04, 2017 and Quanah Gibson-Mount committed Apr 07, 2017

Make two successive modifications of the same attribute separate. This
lets the consumer interpret the log entry the same way as the server
that produced it.

Still depends on the log entry attributes being read in the same order
as they were written.

Howard Chu authored Apr 28, 2016 and Quanah Gibson-Mount committed Oct 17, 2016

Filter needs AttributeAssertions

Howard Chu authored Apr 28, 2016 and Quanah Gibson-Mount committed Oct 17, 2016

and more importantly, avoid escaping requirements that str2filter has

Howard Chu authored Oct 20, 2015 and Quanah Gibson-Mount committed Oct 26, 2015

Windows mutexes are recursive so the trylock test will succeed
even though the mutex is already locked. Just compare be pointers.

Howard Chu authored Jan 30, 2015 and Quanah Gibson-Mount committed Feb 06, 2015

This reduces presentlist memory usage by about 50%. It's still
about 2.5x greater than it should be.

Leonid Yuriev authored Dec 05, 2014 and Quanah Gibson-Mount committed Dec 10, 2014

Mutex must be held even for a read-access of a cookie state,
as there may be a race with write in other thread.