Client: are we using curl efficiently? #4967
Replies: 7 comments
-
I believe we fixed this a little bit differently: #4915 |
Beta Was this translation helpful? Give feedback.
-
I'll give it a try - there are plenty of opportunities at the moment! I'll report back later. |
Beta Was this translation helpful? Give feedback.
-
OK, I've downloaded and am running the artifact. Obviously, I had to shutdown and restart the client, so all the IDs reset to zero. By the time I got to a download page with all files backed off (so I could get a clean log for a single file retry), the result was:
I see that in #4915 @davidpanderson wrote
Anyone help guide me to a similar command to see what (or at least how many) file handles it has open? |
Beta Was this translation helpful? Give feedback.
-
@davidpanderson |
Beta Was this translation helpful? Give feedback.
-
OK, I tried it - will need some help with interpreting it. At the time of taking this log, nine download files were stalled and in various stages of backoff. I also got this warning in terminal:
lsof.log contains 329 lines. |
Beta Was this translation helpful? Give feedback.
-
I would guess the files mentioned by the warnings can be ignored. The warnings appear because the files are on an fuse mount where only that user has access to (not even root). A typical output from LHC@home looks like this:
Here, the mountpoints from the cvmfs client are listed which map some of the CERN repositories into the local directory tree |
Beta Was this translation helpful? Give feedback.
-
Moving this to Discussion since there is no issue currently here. |
Beta Was this translation helpful? Give feedback.
-
The new managers of World Community Grid are struggling to optimize their network connectivity. For me, the major problems are with downloading task datafiles following a successful scheduler contact. The logs sometimes throw up questions worth discussing. Here's the full log of one such download connection.
WCG_download_failure.log
This particular log was captured from BOINC v7.20.2, running under Linux Mint 20.3
Some extracts of interest:
I don't think we make any attempt to capture, interpret and report these useful messages. Should we?
This information message can't be found in client source code - we're assuming it's generated by curl. The difference between oldest and newest connection IDs is some 20k - does our cache really get that big? Should it?
My attention has been drawn to https://curl.se/libcurl/c/CURLMOPT_MAXCONNECTS.html - which suggests that the maximum number of multi-use connections can be limited. Should we use that facility?
Beta Was this translation helpful? Give feedback.
All reactions