Hi, we have a new customer who keeps having node shutdown when they run out of file handles with Kerberos sessions. They raised the ulimit to 262144 files and still run out. They are running v9.1.0-0 and are willing to upgrade if needed. Here is the ticket report for SD02327155:
ertica seems to open a lot of files relating to Kerberos credentials and never closes them, causing a file descriptor leak. After several days, this eventually leads to the process getting close to the max open file limit enforced by the OS (we have it set to 32768).
For example, right now there are zero users connected to the database (apart from the session used to query this):
SELECT count(*) from user_sessions where is_active = true;
count
-------
1
(1 row)
But when I do "lsof" on the process to see what files it has open, I see 666 open files of the form "/var/tmp/krb5....." (many of them are deleted and no longer exist). Although this is small now, the node was restarted a day ago and this number continues to grow until it gets close to the OS limit.
Is there some misconfiguration of our Kerberos settings in Vertica? Or is this a bug where it doesn't close the Kerberos files and leaks the file descriptors?
