read path bug [REPLACEMENT ISSUE]
Closed this issue · 8 comments
The original issue
Id: 21
Title: read path bug
could not be created.
This is a dummy issue, replacing the original one. It contains everything but the original issue description. In case the gitlab repository is still existing, visit the following link to show the original issue:
TODO
In GitLab by @shanedsnyder on May 22, 2018, 16:31
The error always occurs on the read path (though I suppose it could just as easily be a write path bug that isn't noticeable until reading back).
In GitLab by @shanedsnyder on May 22, 2018, 16:31
I'm attaching a patch to mobject-client-test that should demonstrate the bug, in case it's useful.mobject-client-test-4ops.patch
In GitLab by @shanedsnyder on May 22, 2018, 16:36
Note that the error, invalid read of 256 (requested=18446744073709550848) (coming from inside of core-read-op.cpp), suggests that whatever is driving the read request (I guess the segment map?) appears to be supplying a bogus, very large request size (18446744073709550848).
In GitLab by @mdorier on May 22, 2018, 17:35
I'm looking into it.
Looking at the patch: note that you're calling mobject_store_write_op_create(write_op, LIBMOBJECT_CREATE_EXCLUSIVE, NULL); without having initialized the write_op first, and i (which was previously the index of the loop) isn't initialized so it's possible that some junk data is being sent. That being said correcting those I get the same error as you.
In GitLab by @shanedsnyder on May 22, 2018, 18:22
Whoops, that's what happens when I hack the test together too quick. Was basically just trying to mimic an IOR workload I was seeing the bug with. Glad that you can reproduce it, though.
In GitLab by @mdorier on May 22, 2018, 18:47
Bug corrected, you can do a git pull and try now.
In GitLab by @mdorier on May 22, 2018, 18:48
closed
In GitLab by @shanedsnyder on May 23, 2018, 10:58
Works for me :thumbsup_tone1: