[Gluster-devel] Multiplexing status, August 31

[Jeff Darcy]

(This is as much for my own reference as anything, but there seems to be a decent chance that others might be interested)

The good news is that almost all of the tests in tests/basic, except for those related to problematic features (see below) pass now.  Two of the AFR tests fail sporadically because they're timing-dependent.  Self-heal just seems really slow sometimes, so it doesn't complete before the assigned timeout and the tests.  Since I have yet to find any evidence that this is related to multiplexing, investigating this is currently a low priority.  The problematic features fall into two categories:

  Features that can't even coexist with multiplexing yet
    changelog
    trash

  Features that coexist, but most tests still fail
    quota
    snapshot (including USS)
    tiering

The two exceptions that don't have to do with those features are both EC tests, though they're not strictly EC-related.  One test fails because of a problem starting many bricks in quick succession, so that the brick process hasn't fully started before we start pounding it with "attach" requests for the others.  I'm not sure why this only seems to affect one test, but in any event I expect to have that problem fixed soon.  The other problem is that enabling fop statistics doesn't work quite right, because we do it via reconfigure (ick) and that doesn't quite work right with multiple bricks.  That fix will probably take a bit longer.

In addition to those two fixes, and any for the problematic features mentioned above, here are the main development tasks remaining.

* Fix handling of auth options, which currently reside on the server translator and need to be moved (at some point during the startup process) to the bricks below.  This is not currently affecting any tests, but I know it's incorrect.

* Add detection of incompatible transport options (which would preclude multiplexing).  To disable multiplexing, we just act as though all transport options are incompatible.

* Test multiplexing of bricks belonging to different volumes (compatible but with different auth options).

* Measure performance and memory use impacts.


Clearly this is not all going to happen before the end of today for 3.9, but it's looking pretty good for the end of September.  Maybe I'll even be able to show some results at the Berlin summit.