22 August 2017, 14:10 - 16:24
Present: Maiken, Balazs, Anders, David, Aleksandr, Mattias, Oxana
== Issues raised by Maiken
* ARC in PRACE: there's a user who'd like to try ARC and aCT in PRACE, but ARC is not on the list of s/w offered on PRACE; Maiken will investigate how can one make ARC into that list.
* Virtual cluster: Maiken set up a virtual cluster using OpenStack + ElasticCluster at Norwegian academic cloud pilot, physically at USIT, takes ~15 min to set everything up, David will send HammerCloud tests once the certificate isues are solved.
* Site-specific backend patches: Maiken plans to collect ARC back-end patches by sysadmins, will ask people to send such soon. The conclusion was to use SVN for the patches, but Maiken would rather use Gitlab and connected wiki pages.
* SVN -> Git migration: time to make a decision whether to migrate to CodeRefinery's Gitlab or do someting else. Several obstacles are mentioned, along with counter-arguments:
- Nightly builds are not there yet in CodeRefinery's Gitlab; counter-argument: they can be enabled quickly
- Unclear whether merger of various branches for ARC6 should happen first in SVN, or in Gitlab? Maiken suggests to finish development of ARC6 in SVN, and meanwhile migrate ARC5 to Gitlab. When the point comes that we are ready for release we have to evaluate how ready Gitlab is on Coderefinery and how ready are we using it; the condition: having tested tagging and release procedures with ARC5.
- Some argue that Git is not really a revision control system, but a groupware, with a possibility to rewrite history. Counter-argument: these are corner cases, we should not misuse it.
- There are agreed procedures how to use SVN, but none how to use Git, and it will take too much time to agree on how to use Git and to have common understanding. Counter-arguments: we spend more time discussing it, we could have came with agreements already by now; none of those projects that complained about lengthy rule setting switched back to SVN. Aleksandr asks Maiken to prepare the set of working commands, and then we switch to Git in a matter of week and use it in "SVN-style".
- All history may not be preserved. Counter-argument: we don't need much history, information will remain in SVN anyway. Also, we may need several separate repositories, not one single like now.
Some discussions on why documents have a separate release cycle: the main reason was to get rid of LaTeX dependencies.
Conclusion: Maiken should prepare a migration proposal.
The only issue with CodeRefinery is long-term support, which is not certain at the moment. We will deal with it once the problem appears.
New ARC release process will start this week. Question is minor or major? Archery is requested, in addition to python-lrms especially if ARC 6 will not be ready in several months. Maiken sends out email to get opinions on this to NG-core.
== Issue raised by Mattias
It is still unclear whether 5.3.2 is safe to be pushed to EPEL Stable. Everybody forgot what was the remaining issue, memory units? David says that if it is pushed into EPEL, UMD s/w will be broken
== Issues raised by Anders
* A conflict is caused by someone adding python-arc package in Fedora, which has nothing to do with nordugrid-arc. We should ask them to change name since we were there first.
* OpenSSL issues:the new version of OpenSSL 1.1.0 changes behavior of the X509 command-line utility (at least on Fedora 26). The command:
openssl x509 -in $HOME/.globus/usercert.pem -noout -subject
used to produce something like:
subject= /O=Grid/O=NorduGrid/....
With openssl 1.1.0f it now produces:
subject=O = Grid, O = NorduGrid, OU = nbi.dk, CN = Anders Waananen
This has implication for ARC at least for the information system (see revision 32838) and possibly JURA. More investigation about this is needed.
* Some issues on the Debian packaging. Mattias did some cleaning up and improvements which broke older platforms. He will fix it.
== A.O.B.
Digest: write versions consistently (5.3.2 etc)
There are minutes attached to this event.
Show them.
