-Re: Doppleganger threads after ingestion completed
Lance Norskog 2010-06-23, 03:20
I can't think of anything else.
On 6/21/10, [EMAIL PROTECTED] <[EMAIL PROTECTED]> wrote:
> Some answers below.
> (1) netstat -an shows no sockets at all. Remember, the client process is
> gone, dead, shut down.
> (2) This is Solr 1.5 from approximately mid-March.
> (3) Autocommit was on, using the standard configuration present in the
> This could well be a jetty bug and, no, I have not tried tomcat yet.
> From: ext Lance Norskog [[EMAIL PROTECTED]]
> Sent: Sunday, June 20, 2010 10:47 PM
> To: [EMAIL PROTECTED]
> Subject: Re: Doppleganger threads after ingestion completed
> Does 'netstat -an' show incoming sockets for these threads?
> What Solr release is this?
> Is this one long upload of 20m documents without committing? Are you
> doing periodic commits, or automatic commits (in solrconfig.xml)?
> How large are the documents?
> Could this be a jetty bug? Have you tried this on tomcat?
> On Sun, Jun 20, 2010 at 4:50 PM, <[EMAIL PROTECTED]> wrote:
>> So far they've stayed around for 72 hours and counting.
>> Also, I don't care what the stack trace says - CPU is listing as 500%. So
>> it may be momentarily "blocked" but then it must loop.
>> From: ext Lance Norskog [[EMAIL PROTECTED]]
>> Sent: Saturday, June 19, 2010 8:51 PM
>> To: [EMAIL PROTECTED]
>> Subject: Re: Doppleganger threads after ingestion completed
>> "Chewing up cpu" or "blocked". The stack trace says it's blocked.
>> The sockets are abandoned by the program, yes, but TCP/IP itself has a
>> complex sequence for shutting down sockets that takes a few minutes.
>> If these sockets stay around for hours, then there's a real problem.
>> (In fact, there is a bug in the TCP/IP specification, 40 years old,
>> that causes zombie sockets that never shut down.)
>> The HTTP solr server really needs a socket close() method.
>> On Thu, Jun 17, 2010 at 6:08 AM, <[EMAIL PROTECTED]> wrote:
>>> I ran 20,000,000 records into Solr via the extractingUpdateRequestHandler
>>> under jetty. The previous problems with resources have apparently been
>>> resolved by using Http1.1 with keep-alive, rather than creating and
>>> destroying 20,000,000 sockets. ;-) However, after the client terminates,
>>> still find the Solr process chewing away CPU – indeed, there were 5
>>> doing this.
>>> A thread dump yields the following partial trace for all 5 threads:
>>> "btpool0-13" prio=10 tid=0x0000000041391000 nid=0xe7c runnable
>>> java.lang.Thread.State: RUNNABLE
>>> at org.mortbay.jetty.HttpParser$Input.read(HttpParser.java:897)