[SOLR-6110] DIH increase tcp connections so full-import failed

• Type: Bug
• Status: Resolved
• Priority: Blocker
• Resolution: Duplicate
• Affects Version/s: 4.7.2
• Component/s: contrib - DataImportHandler

Environment

Debian Lenny

• Assignee: Unassigned
• Reporter: Gambier
• Created: 22/May/14 14:54
• Updated: 23/May/14 07:42
• Resolved: 23/May/14 07:42

Description

Before i used solr 4.3.1 and i had no problem. Since i use solr 4.7.2 i have a problem with the DIH.
When i do a full import with DIH the number of TCP connection increase.
For example i have 240000 documents to index and the full-import stop randomly between 6000 and 9000 docs indexed with the error :
com.mysql.jdbc.exceptions.jdbc4.CommunicationsException: Communications link failure

With the command (while full-import is running) :
netstat | grep localdo

I saw that there are many connections in status "TIME_WAIT"
The problem is that the OS don't free those connections quickly. So the tcp connections queue increase to its limit and crash ---> full-import failed.

To solve this problem i found a solution :
If i change OS settings "/proc/sys/net/ipv4/tcp_tw_reuse" to 1 and "/proc/sys/net/ipv4/tcp_tw_recycle" to 1 my full-import works to the end.

So why connections are not closed in this new version ?
And why the java socket option SO_REUSEADDR is not used ?

I'm not a Java developer so maybe i miss something but the only way for me to do a success full-import was to use the solution above.

Thank you by advance for your answers to help me solve this problem with a better solution than mine.

Thomas

Issue Links

• duplicates: SOLR-5954 Slower dataimport process caused by not reusing jdbc connections