On Tue, Feb 4, 2020 at 10:03 AM Thomas Munro <[email protected]> wrote: > On Tue, Feb 4, 2020 at 9:13 AM Peter Geoghegan <[email protected]> wrote: > > Uh, this seems to have completely disabled parallel index builds on > > the master branch. > > Oops. The check needs to move down below InitializeParallelDSM(), and > release any extra resources that might have been acquired (snapshot?). > I will do some testing in a few hours and post a fix.
Here is take 2. I tested CI and CIC, and verified that workers are started or not depending on dsm_create() failure, using the attached fault injector patch. (Fuzzing the regression tests repeatedly with that applied also revealed similar problems elsewhere in the tree, which I'll write about in a new thread.)
0001-Handle-lack-of-DSM-slots-in-parallel-btree-build-tak.patch
Description: Binary data
0002-Inject-random-faults-into-dsm_create.patch
Description: Binary data
