Received-SPF: neutral (sog-mx-3.v43.ch3.sourceforge.com: 67.222.55.9 is
	neither permitted nor denied by domain of trillion01.com)
	client-ip=67.222.55.9; envelope-from=olivier@trillion01.com;
	helo=oproxy7-pub.mail.unifiedlayer.com; 
Message-ID: <1380600219.932.21.camel@Wailaba2>
From: Olivier Langlois <olivier@trillion01.com>
To: slush <slush@centrum.cz>
Date: Tue, 01 Oct 2013 00:03:39 -0400
In-Reply-To: <CAJna-Hi+eyRnZUtHpfvod_uRCmjPOL5HS3ZZpr54yzbKRRT9-w@mail.gmail.com>
References: <CAJna-Hi+eyRnZUtHpfvod_uRCmjPOL5HS3ZZpr54yzbKRRT9-w@mail.gmail.com>
Organization: Trillion01 Inc
Content-Type: text/plain; charset="ISO-8859-1"
Mime-Version: 1.0
Content-Transfer-Encoding: 7bit
Cc: "bitcoin-development@lists.sourceforge.net"
	<bitcoin-development@lists.sourceforge.net>
Subject: Re: [Bitcoin-development] bitcoind stops responding
Precedence: list

On Mon, 2013-09-30 at 22:44 +0200, slush wrote:
> Hi,
> 
> 
> during several weeks I'm observing more and more frequent issues with
> bitcoind. The problem is that bitcoind stops responding to RPC calls,
> but there's no other suspicious activity in bitcoind log, CPU usage is
> low, disk I/O is standard etc.
> 
> 
> I observed this problem with version 0.8.2, but it is still happening
> with 0.8.5. Originally this happen just one or twice per day. Today my
> monitoring scripts restarted bitcoind more than 30x, which sounds
> alarming. This happen on various backends, so it isn't a problem of
> one specific node. Is there anybody else who's observing similar
> problem? 

What a coincidence. I do have observed the same thing. right now with
0.8.5. I am writing a small app. My jsonrpc client is programmed to
timeout after 2 secs and I did see a couple of timeouts once in while.

What I did is a simple test app that just hammer bitcoind with 3 rpc
requests every 30 seconds and I abort it as soon as it encountered a
timeout.

The 3 request burst is performed on the same HTTP 1.1 kept alive
connection. Then I disconnect. When I launch my app before leaving in
the morning, pretty sure that I have a core dump waiting for me when I
come back.

I choose very simple calls: getinfo,getaccount

Added a couple of traces in the RPC handling code. (BTW, timestamps in
traces would be tremendously useful for tracking problems...). I see my
request received by bitcoind but there is no trace yet to show that the
reply is sent.

Not sure yet exactly where the problem is but my current #1 suspect is:

LOCK2(cs_main, pwalletMain->cs_wallet);

with some kind of lock contention with the other threads.