KIRTI RUGE created HIVE-28491:
---------------------------------
Summary: Check feasibility of improvement in
ShowTableStatusOperation.
Key: HIVE-28491
URL: https://issues.apache.org/jira/browse/HIVE-28491
Project: Hive
Issue Type: Improvement
Components: Hive
Reporter: KIRTI RUGE
SHOW TABLE EXTENDED IN test_db LIKE foo_n4; takes around 25 minutes to 2.5
hours on live cluster when table has huge partitions
Scenario given:
# A table has 8.8k partitions and 26k files+dirs in the table location.
# From production jstack it is clearly shown that method
writeFileSystemStats() is calling getfilestatus and liststatus for every single
file and dir and table location of given table.
# ShowTableStatusOperation->TextMetaDataFormatter. showTableStatus is a single
call which tried to fetch fileStatus for all paths of partitions of given table.
# Check feasibility of improvement of above code if it can be be optimized
with more elegant code like-
instead of going over fs.getFileStatus(loc) for each location we can think of
using RemoteIterator
like RemoteIterator<LocatedFileStatus> remoteIterator =
fs.listFiles(base.getPath(), ifneededrecursive:true);
while (remoteIterator.hasNext())
{ LocatedFileStatus each = remoteIterator.next(); each.getLen();
each.getAccessTime(); each.getModificationTime() }
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
