When checking something else in the base backup code, I've noticed that
sendFileWithContent() does not advance the 'content' pointer. The sink buffer
is large enough (32kB) so that the first iteration usually processes the whole
file (only special files are processed by this function), and thus that the
problem is hidden.
However it's possible to hit the issue: if there are too many tablespaces,
pg_basebackup generates corrupted tablespace_map. Instead of writing all the
tablespace paths it writes only some and then starts to write the contents
from the beginning again.
The attached script generates scripts to create many tablespaces as well as
the underlying directories. Fix is attached here as well.
--
Antonin Houska
Web: https://www.cybertec-postgresql.com
#!/bin/bash
TBSPDIR=/mnt/ramdisk/tbspcs
TBSPCOUNT=2048
SCRIPT_SH=create.sh
SCRIPT_SQL=create.sql
echo "#!/bin/bash" > ${SCRIPT_SH}
echo "mkdir -p $TBSPDIR" >> ${SCRIPT_SH}
echo "" > ${SCRIPT_SQL}
i=0
while [ $i -lt $TBSPCOUNT ];
do
printf "mkdir %s/%.4d\n" $TBSPDIR $i >> ${SCRIPT_SH}
printf "CREATE TABLESPACE tbsp_%.4d LOCATION '%s/%.4d';\n" $i $TBSPDIR $i >> ${SCRIPT_SQL}
i=$((i+1))
done
chmod u+x ${SCRIPT_SH}
diff --git a/src/backend/backup/basebackup.c b/src/backend/backup/basebackup.c
index 74fb529380..7aa5f6e44d 100644
--- a/src/backend/backup/basebackup.c
+++ b/src/backend/backup/basebackup.c
@@ -1073,6 +1073,7 @@ sendFileWithContent(bbsink *sink, const char *filename, const char *content,
memcpy(sink->bbs_buffer, content, nbytes);
bbsink_archive_contents(sink, nbytes);
bytes_done += nbytes;
+ content += nbytes;
}
_tarWritePadding(sink, len);
