Open fulara opened 4 years ago
Hi @fulara
The parallel download and the parallel upload are two very different features, implement in different places with different code. Please use separate issue for both.
Let's use this issue for the download one, if you see errors in the upload, please open a new one. We need to start narrowing the causes. First thing is that build --missing
should be totally irrelevant, because the downloads are done before the builds. Is this issue reproducible with just some "hello world" simple packages? I think it would be good to try to isolate from the environment. Can you reproduce somehow also without running in the bamboo CI?
Hello @memsharded okay. Yea - I am seeing issues with both dl and upload: see my no4 issue:
ERROR: stats/3.1:f1c30f8929c64350fcc4683162cfbebc5bdf34c0: Upload package to 'ig' failed: 'NVM_NODEJS_ORG_MIRROR'. [Remote: ig]
After weekend I'll be able to toy around with this, yes, ofc build --missing
is irrelevant here. I'll get back to this as soon as i can.
I'll keep this for download
only.
So i've done some tests. Somehow i was initially lucky with reproducing it but then a lot of no lucks. Anyway this is the error I was getting today, invoking build manually ad not out of bamboo: Got it three times out of 40 builds. Originally when I created this issue i was almost getting 1/1
ERROR: Download failed, check server, possibly try again
attempt to release recursive lock not owned by thread
I got that again on my complex package with code and deps etc.
Then i went ahead and generated a tree of dummy conans so something like generate 100 levels of conanfile that each next depend on previous one, I didnt get that 'attempt to release recursive lock' error even once during ~100 downloads.
I was sometimes getting the CERTIFICATE error from above..
I was testing with packages weighing 10mb and 100kb,
Is there any logging you'd want enabled for this certificate or 'attempt to release' lock thing? When I get some more time i'll try to change this to instead work on some kind of buildable projects but not sure if that woul dmake any difference.
It seems that such assertion comes from the tqdm
library, and our progress bars that are not properly managed:
https://github.com/iterative/dvc/issues/2589#issuecomment-543823689
Also it could be related to the installer, if it is pyinstaller or not.
It would be useful to know:
Updated original comment but same details here as well: also added conan.conf we are using.
OS: built rh6 ran on rh6 and rh7. installed: pyinstaller.
I am worried that certificate thing is a different thing though.
Is this issue on the artifactory side rather than conan ? I dont know. I have enabled parallel download and upload, on both of these we are getting random failures.
To provide more inforamtion, because some of the errors refers to certificates we provide a custom certificate that usually seems to be working.
Environment Details (include every applicable attribute)
How do we provide conan
We bundle conan using pyinstaller, on rh6 machine roughly following this procedure here: https://github.com/conan-io/conan/issues/6325 We then distribute that pyinstalled package across and just use it as is.
Do we use download cache?
No, we download everything from scratch everytime conan.conf:
Steps to reproduce (Include if Applicable)
Have a package with quite a few dependencies and do
conan create .
andupload *
to force loads of uploads and downloads.Exact procedure we are doing.
Before each build we are downloading are bundled conan package and and initializing this from scratch, and then just build particular package. We have provided custom
cacert.pem
I am getting various set of errors:
Logs (Executed commands with output) (Include/Attach if Applicable)
Error no1 ConanException - MAIL
Error no2 ceritificate spurious failure
Error no3 random bamboo variable?
Error no4 some npm error?
I have run around 40 builds without parallel settings and this works then without issue.
references: https://github.com/conan-io/conan/pull/6632 https://github.com/conan-io/conan/pull/5856