Fix possible failure to mark hash metapage dirty.

Report and suggested fix by Lixian Zou.  Amit Kapila put it
in the form of a patch and reviewed.

Discussion: http://postgr.es/m/151739848647.1239.12528851873396651946@wrigleys.postgresql.org

diff --git a/src/backend/access/hash/hashovfl.c b/src/backend/access/hash/hashovfl.c
index c206e704d4d1d326dd4429421b26e5acb15c7013..912e38ca71e47f3d665b3cb4c4538501b49a7812 100644 (file)
--- a/src/backend/access/hash/hashovfl.c
+++ b/src/backend/access/hash/hashovfl.c
@@ -341,9 +341,10 @@ found:
            metap->hashm_mapp[metap->hashm_nmaps] = BufferGetBlockNumber(newmapbuf);
            metap->hashm_nmaps++;
            metap->hashm_spares[splitnum]++;
-           MarkBufferDirty(metabuf);
        }
 
+       MarkBufferDirty(metabuf);
+
        /*
         * for new overflow page, we don't need to explicitly set the bit in
         * bitmap page, as by default that will be set to "in use".