On Sun, Dec 31, 2017 at 1:50 PM, Ibrahim Edib Kokdemir
<kokdemir@gmail.com> wrote:> * write_cache is disabled
> * there is no incorrect work_mem parameter setting.
> * logical dump is working, (maybe) no curruption in data.
> * there is streaming replication, we do not repeat the error in the
> replicas. (replicas in different minor versions, 9.6.4, 9.6.3 accordingly)
> * we have large_object field, logical_dump also works with large_objects
> fields.
>
> Any idea?
This is very likely to be corruption. It's important to determine the
cause and extent of this corruption. I suggest using amcheck for this,
which is available for those Postgres versions from:
https://github.com/petergeoghegan/amcheck
Note that there are Debian and Redhat packages available.
You'll definitely want to use the "heapallindexed" option here, at
least for primary key indexes (pass "pg_index.indisprimary" as
"heapallindexed" argument, while generalizing from the example SQL
query for bt_index_check()). This process has a good chance of
isolating the problem, especially if you let this list see any errors
raised by the tool.
--
Peter Geoghegan