Details
Description
Restore snapshot makes parent split region online as shown in the attached snapshot.
Steps to reproduce
=====================
1. Create table
2. Insert few records into the table
3. flush the table
4. Split the table
5. Create snapshot before catalog janitor clears the parent region entry from meta.
6. Restore snapshot
We can see the problem in meta entries,
Meta content before restore snapshot:
t1,,1511537529449.077a12b0b3c91b053fa95223635f9543. column=info:regioninfo, timestamp=1511537565964, value={ENCODED => 077a12b0b3c91b053fa95223635f9543, NAME => 't1,,1511537529449.077a12b0b3c91b053fa95223635f9543.', STARTKEY => '', ENDKEY => '', OFFLINE => true, SPLIT => true} t1,,1511537529449.077a12b0b3c91b053fa95223635f9543. column=info:seqnumDuringOpen, timestamp=1511537530107, value=\x00\x00\x00\x00\x00\x00\x00\x02 t1,,1511537529449.077a12b0b3c91b053fa95223635f9543. column=info:server, timestamp=1511537530107, value=host-xx:16020 t1,,1511537529449.077a12b0b3c91b053fa95223635f9543. column=info:serverstartcode, timestamp=1511537530107, value=1511537511523 t1,,1511537529449.077a12b0b3c91b053fa95223635f9543. column=info:splitA, timestamp=1511537565964, value={ENCODED => 3c7c866d4df370c586131a4cbe0ef6a8, NAME => 't1,,1511537565718.3c7c866d4df370c586131a4cbe0ef6a8.', STARTKEY => '', ENDKEY => 'm'} t1,,1511537529449.077a12b0b3c91b053fa95223635f9543. column=info:splitB, timestamp=1511537565964, value={ENCODED => dc7facd824c85b94e5bf6a2e6b5f5efc, NAME => 't1,m,1511537565718.dc7facd824c85b94e5bf6a2e6b5f5efc.', STARTKEY => 'm ', ENDKEY => ''} t1,,1511537565718.3c7c866d4df370c586131a4cbe0ef6a8. column=info:regioninfo, timestamp=1511537566075, value={ENCODED => 3c7c866d4df370c586131a4cbe0ef6a8, NAME => 't1,,1511537565718.3c7c866d4df370c586131a4cbe0ef6a8.', STARTKEY => '', ENDKEY => 'm'} t1,,1511537565718.3c7c866d4df370c586131a4cbe0ef6a8. column=info:seqnumDuringOpen, timestamp=1511537566075, value=\x00\x00\x00\x00\x00\x00\x00\x02 t1,,1511537565718.3c7c866d4df370c586131a4cbe0ef6a8. column=info:server, timestamp=1511537566075, value=host-xx:16020 t1,,1511537565718.3c7c866d4df370c586131a4cbe0ef6a8. column=info:serverstartcode, timestamp=1511537566075, value=1511537511523 t1,m,1511537565718.dc7facd824c85b94e5bf6a2e6b5f5efc. column=info:regioninfo, timestamp=1511537566069, value={ENCODED => dc7facd824c85b94e5bf6a2e6b5f5efc, NAME => 't1,m,1511537565718.dc7facd824c85b94e5bf6a2e6b5f5efc.', STARTKEY = > 'm', ENDKEY => ''} t1,m,1511537565718.dc7facd824c85b94e5bf6a2e6b5f5efc. column=info:seqnumDuringOpen, timestamp=1511537566069, value=\x00\x00\x00\x00\x00\x00\x00\x08 t1,m,1511537565718.dc7facd824c85b94e5bf6a2e6b5f5efc. column=info:server, timestamp=1511537566069, value=host-xx:16020 t1,m,1511537565718.dc7facd824c85b94e5bf6a2e6b5f5efc. column=info:serverstartcode, timestamp=1511537566069, value=1511537511523
Meta content after restore snapshot:
t1,,1511537529449.077a12b0b3c91b053fa95223635f9543. column=info:regioninfo, timestamp=1511537667635, value={ENCODED => 077a12b0b3c91b053fa95223635f9543, NAME => 't1,,1511537529449.077a12b0b3c91b053fa95223635f9543.', STARTKEY => '', ENDKEY => ''} t1,,1511537529449.077a12b0b3c91b053fa95223635f9543. column=info:seqnumDuringOpen, timestamp=1511537667635, value=\x00\x00\x00\x00\x00\x00\x00\x0A t1,,1511537529449.077a12b0b3c91b053fa95223635f9543. column=info:server, timestamp=1511537667635, value=host-xx:16020 t1,,1511537529449.077a12b0b3c91b053fa95223635f9543. column=info:serverstartcode, timestamp=1511537667635, value=1511537511523 t1,,1511537565718.3c7c866d4df370c586131a4cbe0ef6a8. column=info:regioninfo, timestamp=1511537667598, value={ENCODED => 3c7c866d4df370c586131a4cbe0ef6a8, NAME => 't1,,1511537565718.3c7c866d4df370c586131a4cbe0ef6a8.', STARTKEY => '', ENDKEY => 'm'} t1,,1511537565718.3c7c866d4df370c586131a4cbe0ef6a8. column=info:seqnumDuringOpen, timestamp=1511537667598, value=\x00\x00\x00\x00\x00\x00\x00\x0B t1,,1511537565718.3c7c866d4df370c586131a4cbe0ef6a8. column=info:server, timestamp=1511537667598, value=host-xx:16020 t1,,1511537565718.3c7c866d4df370c586131a4cbe0ef6a8. column=info:serverstartcode, timestamp=1511537667598, value=1511537511523 t1,m,1511537565718.dc7facd824c85b94e5bf6a2e6b5f5efc. column=info:regioninfo, timestamp=1511537667621, value={ENCODED => dc7facd824c85b94e5bf6a2e6b5f5efc, NAME => 't1,m,1511537565718.dc7facd824c85b94e5bf6a2e6b5f5efc.', STARTKEY = > 'm', ENDKEY => ''} t1,m,1511537565718.dc7facd824c85b94e5bf6a2e6b5f5efc. column=info:seqnumDuringOpen, timestamp=1511537667621, value=\x00\x00\x00\x00\x00\x00\x00\x0D t1,m,1511537565718.dc7facd824c85b94e5bf6a2e6b5f5efc. column=info:server, timestamp=1511537667621, value=host-xx:16020 t1,m,1511537565718.dc7facd824c85b94e5bf6a2e6b5f5efc. column=info:serverstartcode, timestamp=1511537667621, value=1511537511523
Root Cause:
We dont update the region split information in .regioninfo file in HDFS, but while restoring the snapshot we set regioninfo based on the .regioninfo entries,
// Identify which region are still available and which not. // NOTE: we rely upon the region name as: "table name, start key, end key" List<HRegionInfo> tableRegions = getTableRegions(); if (tableRegions != null) { monitor.rethrowException(); for (HRegionInfo regionInfo: tableRegions) { String regionName = regionInfo.getEncodedName(); if (regionNames.contains(regionName)) { LOG.info("region to restore: " + regionName); regionNames.remove(regionName); metaChanges.addRegionToRestore(regionInfo); } else { LOG.info("region to remove: " + regionName); metaChanges.addRegionToRemove(regionInfo); } }
Here getTableRegions() is retrieved from HDFS.
There can be two solutions,
1. Set the regioninfo based on the snapshot-manifest details.
2. Update the .regioninfo after region split