On Fri, 2016-03-11 at 17:24 +0100, Michael Paquier wrote:
> On Fri, Mar 11, 2016 at 5:19 PM, Anastasia Lubennikova wrote:
> >
> > BTW, if you know a good way to corrupt index (and do it
> > reproducible) I'd be
> > very glad to see it.
> You can use for example dd in non-truncate mode to corrupt on-disk
> page data, say that for example:
> dd if=/dev/random bs=8192 count=1 \
> seek=$BLOCK_ID of=base/$DBOID/$RELFILENODE \
> conv=notrunc
I guess /dev/random is not very compatible with the "reproducible"
requirement. I mean, it will reproducibly break the page, but pretty
much completely, which is mostly what checksums are for.
You can actually pretty easily produce a test case by setting up streaming replication between servers running two different version of glibc.
I actually wrote a tool that spins up a pair of VMs using vagrant and then sets them up as streaming replica's using ansible. It provides a nice one liner to get a streaming replica test environment going and it will easily provide the cross glibc test case. Technically, though it belongs to Trip because I wrote it on company time. Let me see if I can open source a version of it later this week that way you can use it for testing.
- Matt K.