|
| 1 | +--- |
| 2 | +layout: article |
| 3 | +title: Summary of 2th Quattor workshop (2018-10-29 to 2018-10-31, University of Gent) |
| 4 | +category: meeting |
| 5 | +author: Michel Jouvin |
| 6 | +--- |
| 7 | + |
| 8 | +# Quattor Workshop - UGent - 29-31/10/2018 |
| 9 | + |
| 10 | + |
| 11 | +## Site Reports |
| 12 | + |
| 13 | +### UGent |
| 14 | + |
| 15 | +Considering looking at Aquilon but for the time being still will SCDB |
| 16 | +* Tweaked it more in the meantime... |
| 17 | + |
| 18 | +OpenStack: wrote a component that can manage/configure most of OpenStack services |
| 19 | + |
| 20 | +### RAL |
| 21 | + |
| 22 | +More and more Quattor/Aquilon usage |
| 23 | +* HPC on board |
| 24 | +* Other system groups suggested to use it: want to promote commonality across the department |
| 25 | + |
| 26 | +Aquilon: working well but with a few local tweaks |
| 27 | + |
| 28 | +Ceph: less and less configuration to do at each new release, easier to support in Quattor |
| 29 | + |
| 30 | +Work in progress with FreeIPA but still a long journey... |
| 31 | +* Need to cleanup LDAP and NIS services first... |
| 32 | + |
| 33 | +James now more involved in HTC configuration/management |
| 34 | +* Batch system: HTCondor |
| 35 | +* Also managing the teams in charge of grid GOCDB and APEL |
| 36 | + |
| 37 | +### MS |
| 38 | + |
| 39 | +Big challenge: integrate new teams, teach them best practices |
| 40 | +* Code review is a key (Gerrit) |
| 41 | +* panlint helps |
| 42 | +* More documentation on how to right proper pan would help |
| 43 | + |
| 44 | +Looking at integrating Docker/Kubernetes as the provisioning infrastructure for grid |
| 45 | + |
| 46 | +Move out of AFS impacts Kerberos: which alternatives? FreeIPA? Active Directory? |
| 47 | + |
| 48 | +Still has to support `RHEL5`: impacts ability to upgrade AII servers |
| 49 | +* Plan to look at `RHEL8` impact on Quattor when it is available: hopefully nothing disruptive (based on Fedora)... |
| 50 | + |
| 51 | +Working on pushing upstream MS modifications to components. |
| 52 | + |
| 53 | +### LAL |
| 54 | + |
| 55 | +Still running SCDB but Aquilon production instance started! |
| 56 | +* Manages a few nodes: need time to review/migrate site templates |
| 57 | +* Using Active Directory as the Kerberos infrastructure |
| 58 | +* Deployment server shared with SCDB |
| 59 | +* Very impressed by Aquilon quality |
| 60 | +* Documentation updated on quattor.org: should be pretty complete to start |
| 61 | + * A preliminary SCDB migration guide available |
| 62 | + |
| 63 | +Aquilon: several contributions... and open issues |
| 64 | +* Good collaboration with MS after initial definition of the process |
| 65 | + |
| 66 | +Open question: how to model/integrate the cloud infrastructure in Aquilon |
| 67 | +* Don't want to track in Aquilon what is tracked in the cloud, like VM placement |
| 68 | +* The OpenStack cloud is our base infrastructure: more and more virtualised services with system image |
| 69 | +managed with Quattor |
| 70 | + * Service data in a persistent disk |
| 71 | + * Image built from a base image + Quattor customisation |
| 72 | + |
| 73 | +Another wish for Aquilon: be able to define site-specific key/value pairs in Aquilon that will be passed to |
| 74 | +the plenary templates in some ways |
| 75 | + |
| 76 | + |
| 77 | +## Debian Support |
| 78 | + |
| 79 | +No progress since last workshop. Basically working except the network configuration. |
| 80 | +* Debian packaging done from RPM after unpacking them: hacky at this time, limited functionality (no pre and post script support) |
| 81 | + * Files are properly relocated to be at the standard place in Debian |
| 82 | +* No initial install support |
| 83 | + * Debian said to support Kickstart files: is it worth exploring? |
| 84 | + * RAL focus: management of (CumulusOS) switches where initial install is not an issue. |
| 85 | + * Same use case at UGent: currently using `metaconfig` to generate the appropriate config file and push them manually |
| 86 | + to the switch |
| 87 | + |
| 88 | +Main issue currently: a bug in `ncm-spma` apt back-end leading to packages not installed or with the wrong version. |
| 89 | +* See https://github.com/quattor/configuration-modules-core/pull/1258 |
| 90 | + |
| 91 | + |
| 92 | +## OpenStack Support |
| 93 | + |
| 94 | +2 approaches currently for OpenStack configuration |
| 95 | +* The template library based on `metaconfig` to produce the required configuration |
| 96 | + * Makes easy to support different OpenStack version with a different branch of the templates, like it was done for |
| 97 | + grid middleware |
| 98 | + * Used by RAL, LAL and Strasbourg |
| 99 | +* `ncm-openstack` by UGent that does everything: used to support `Ocata`, now supports Pike |
| 100 | + * Internally based on `CAFRender` and Template Toolkit files: makes easy to add a new service (highly based on `metaconfig` code) |
| 101 | + * The component is sensitive to OpenStack configuration changes with new versions: often significant changes |
| 102 | + * Endpoints managed by Quattor |
| 103 | + * Keystone users for the OpenStack services managed by Quattor |
| 104 | + |
| 105 | +It would be good to see if the configuration module could be used with the template library |
| 106 | +* Main challenge is the support of multiple versions in the configuration module: impact on the schema and Template Toolkit files |
| 107 | +* Alex interested to have a look |
| 108 | + |
| 109 | +Work in progress at UGent on OpenID support but it will remain necessary to create users manually in OpenStack |
| 110 | + |
| 111 | +Also started with Manilla: some issues related to the fact that the NFS server needs to be managed outside OpenStack |
| 112 | +* Need to ensure that network routing is properly configured from the VM network to the NFS server |
| 113 | +* UGent: NFS server is based on Ganesha in front of Ceph |
| 114 | + * No performance figures yet |
| 115 | + |
| 116 | + |
| 117 | +## Quattor Releases |
| 118 | + |
| 119 | +Next one planned: 18.12 |
| 120 | +* By Christmas |
| 121 | +* 3 releases a year seems a good rhythm |
| 122 | + * 18.9 will not happen: not enough availability during summer |
| 123 | + |
| 124 | +New layout for template library repositories: need to experiment with several ideas first, no planned soon... |
| 125 | +* See https://github.com/quattor/release/issues/309 |
| 126 | +* Also needed to configuration modules |
| 127 | +* Approach foreseen: pull request from the source repository to the destination one, possibly after moving a branch content |
| 128 | + in a directory (in the source repository?) |
| 129 | +* Also need to move open issues and pull requests to new repository (will not preserve numbers) |
| 130 | + * A tool exists at GitHub for doing it (with cross references in both repositories) |
0 commit comments