6.60b15 Error Header DL Failed: ReadFailure and memleak

[syshog]

I've had this error

[21:53:13] ERROR AW alt.binaries.boneless Header Download Failed: ReadFailure

A few times when upgrading from 6.56 to 6.60 beta 15. I've used two different backups of headers each batch of headers have updated successfully when tested with 6.56. The error occurs after all header updates have been downloaded and only on the last group being processed by NB. I use the verbose logging so I'm not sure if anybody else that downloads headers necessarily sees this error message. The lines that appear before the fail message are the standard compress=gzip for each group being processed before it reaches the last group and is the last line in the log, the group name changes depending on the last group those headers were downloaded.

As a side note there's still a memory leak when loading headers, I tested it with loading all headers from a.b.warez and a.b.boneless, in most cases NB's memory usage matched what was reported in task manager in one out three times the system would report 10GB more memory usage than NB was reporting once all the headers were loaded and accessible. The tests in this case were done with NB running with firefox, antivirus (with exceptions for the Drive containing NB plus configurations in place), and a firewall running.

Has anyone else experienced these problems?

[Quade]

1 - It's just your typical network error. Newsbin will re-connect and continue.


2 -

would report 10GB more memory usage than NB was reporting once all the headers were loaded and accessible.

Don't know what this means. Newsbin doesn't report memory usage.

I did some fairly extensive testing last week and couldn't ID a memory leak. I did notice that Windows would often report more memory usage in Newsbin than it was actually using. I could cause Windows to flush the memory out by loading 2 gigs of memory worth of headers, then closing the tab. The would return Newsbin's reported memory usage to what I'd expect with no tabs open. I just chalked this up to inscrutable windows memory managements beyond my control.

[syshog]

1) I'm experiencing the problem quite often with beta 15 and it's reproducible on my end. It acts like the previous attempt to grab headers completed and the data was imported in the db but when loading headers there are massive amount of incompletes. If I used 6.56 and update headers I do not run into this problem.

2)I used task manager to look at per process memory usage. A 2GB test is not a reliable memory leak test, NB has always displayed the problem more frequently above 10GB of memory for myself. boneless and warez have always been my benchmark for displaying NB's memleak problem, boneless will use 27GB ram fully and 1 out of 3 times windows will report 38GB of ram being used instead of 29GB, about a 1GB of ram tends to be used for cache, warez will result in 24GB of ram in usage, but in 1 out of 3 attempts it will spike to 34GB of usage. Now with 6.56 I have a little higher mem usage because your tweaks from the 6.60 betas are not present.

[Quade]

, boneless will use 27GB ram fully and 1 out of 3 times windows will report 38GB of ram being used instead of 29GB, about a 1GB of ram tends to be used for cache, warez will result in 24GB of ram in usage, but in 1 out of 3 attempts it will spike to 34GB of usage. Now with 6.56 I have a little higher mem usage because your tweaks from the 6.60 betas are not present.

I don't believe it's leaking any ram. I just don't think you're getting accurate numbers off task manager. That was my point with my comment about loading up 2 gigs forcing windows to flush the rest and report the accurate number. I only use the 32 bit version but the code is identical between them. You must have a high chunk cache for it to be using 1 Gig for caching.