I did manage to trace the cause of the "epoll_ctl(EPOLL_CTL_DEL): failed on fd=197" messages. The latest epoll patch fixes this problem. There are 2 messages that the new patch will cause to appear with the default debug options (ALL,1): storeClientCopy3: <url> - clearing ENTRY_DEFER_READ This is caused when an object has been deferred, and for some reason the memory has not been freed, even though all clients have seen all in-memory parts of the url. I can't see why this would be the case, and I'm sure it's not caused by the epoll code, but the epoll code needs to catch this condition. WARNING defer handler for fd=<fd>(<url>) does not call commDeferFD() - backing off manually. This is caused when an object has data ready to be read, and the defer handler reports that the read should be deferred, but does not tell the epoll code to back off. This debug would probably be very noisy if used with delay pools, but other than that it indicates a non-optimised defer handler. These two debug statements do not produce a large number of entries in the debug logs (181 and 52 messages respectively out of 3204 messages total in one of our proxy server's cache.log out of a total of ~4.2 million requests). The proxy server does handle these cases appropriately, but they indicate sub-optimal performance. As they represent a tiny fraction of the total number of requests, I have not spent any time working out the exact cause of these problems. Regards Steven > -----Original Message----- > From: Sally Huang [mailto:sallyhsl@xxxxxxxxx] > Sent: Wednesday, May 11, 2005 9:34 AM > To: Henrik Nordstrom > Cc: squid-users@xxxxxxxxxxxxxxx > Subject: Re: Odp: Re: [squid-users] how to apply epoll-2_5 > patch to squid2.5-stable9 > > Thanks for your clarify. > > The author doesn't reply me. Could you pls do me a favor and check > with the author whether he has updated the latest epoll patch so that > everyone won't face this "epoll_ctl(EPOLL_CTL_DEL): failed on fd=197" > problem? > > Regards, > sally > > > On 5/10/05, Henrik Nordstrom <hno@xxxxxxxxxxxxxxx> wrote: > > > > > > On Tue, 10 May 2005, Sally Huang wrote: > > > > > Do you mean the " epoll_ctl(EPOLL_CTL_DEL): failed on fd=197" bug > > > isalready fixed on both epoll-2_5 patch and epoll-2.5 branch in > > > squidCVS? > > > > I am the wrong person to answer that question, but as you I > remember the > > branch author mentioning that there has been fixes in that area. > > > > I answerd to the question how you could get the latest > version of the > > epoll branch. > > > > Regards > > Henrik > > > > -- > No virus found in this incoming message. > Checked by AVG Anti-Virus. > Version: 7.0.308 / Virus Database: 266.11.8 - Release Date: 5/10/2005 > > -- No virus found in this outgoing message. Checked by AVG Anti-Virus. Version: 7.0.308 / Virus Database: 266.11.8 - Release Date: 5/10/2005
