[top-posting to retain the quoted mail in entirety]
Thank you Toshio for the detailed information. This does provide some clarity about the events in the background that eventually led to the decision. Since we are already into the transition phase, further debate on the 'hows' and 'whys' and 'so thats what it really was' would not serve much purpose besides adding to the noise.
However, since we do expect to review the decision and explore the possibility of returning to a Fedora hosted infrastructure, we may as well ensure that we do not repeat the mistakes, especially that of communicating. As opposed to the individuals from the FLP, the FLSCo, an elected group of representatives from the FLP did not seem to figure anywhere during all this (unreported) discussion. As a result, an established communication channel within the FLP was completely ignored or not utilized.
The 'threat board' is perhaps another good thing to have in the future. With red signs flashing all over the FLSCo and FLP could have pushed for discussion much earlier. It would also allow individual project teams to check back on the general health of the tools that they use.
Thank you for your patience and hopefully we can see a much stable translation infrastructure in place for the FLP.
regards Runa
On শুক্রবার 25 ফেব্রুয়ারি 2011 06:14 , Toshio Kuratomi wrote:
On Thu, Feb 24, 2011 at 02:08:40PM -0800, Karsten Wade wrote:
Definitely there was some process breakdown and some lack-of-process.
It seems that it wasn't an active open community discussion between Infrastructure and affected teams about problems with maintaining translate.fp.org and how to fix them. This all could have been discussed over the preceding months, presuming that were possible ... and I am going to presume it wasn't, that we are here based on a series of best-efforts and best-will from people involved. Does that make sense?
So... you'll need to contact mmcgrath for exactly where, with whom, and when previous discussions took place but discussion had been taking place for quite a while. A lot of people realized there was an issue when this ticket took 9 months to resolve: https://fedorahosted.org/fedora-infrastructure/ticket/1455
Enough that gomix, beckerde, other translators, and I discussed the situation at FUDCon Chile last year and I opened this ticket: https://fedorahosted.org/fedora-infrastructure/ticket/2277
The first meeting that we (infra) discussed that ticket at, mmcgrath said that he'd tried but failed to find someone to maintain transifex in infra from the l10n team. He and I tried to think of someone that we knew personally that could do the job and via email got in touch with beckerde who agreed to work on packaging and migration of the new transifex.
Since beckerde was new to Fedora packaging and transifex dependencies had changed it took a while to get things working but with the help of other latam packagers we got packaging sorted out eventually and started to test upgrades on the infrastructure staging servers. The version that's being worked on is an older version because newer versions of transifex would change the translation workflow (the recent talks have made clear that the change in workflow is mainly for the developers of the translated software rather than the translators.)
{{A side thread to this story: In the F14 time frame (ticket 1455 time frame), the translation teams made some valid noise about the problems that the version of transifex we were running were causing them. At that time, glezos, stickster, some people from l10n and infra analyzed the possibility of moving to transifex.net and decided that it was possible but were divided about the desirability. We were able to upgrade the FI instance of transifex which deferred the question.}}
Sometime between the tests on stg and fudcon some of infra starts worrying about how we're going to continue maintaining transifex. beckerde is signing up to work on packaging the software and deploying it but we're still running several versions behind and we don't have in-house knowledge on how to fix issues in the code. Moving to the newest version is caught in the workflow changing problem that, since infra isn't involved in using transifex, none of us know what is changing or how to approach l10n about changing their workflow to accomodate it.
{{Next side thread: mmcgrath leaves as infrastructure leader. This leaves us shorthanded as we were already running close to capacity and mmcgrath was also a good community leader who was able to entice and induct volunteers.}}
At FUDCon, smooge, skvidal, ricky, CodeBlock meet with igorps and some other translators that are present and talk about moving off of Fedora hosting transifex if glezos still wants to host us. We talk about a gradual shift at that time. Later, infrastructure members talk to glezos on IRC and more with each other.
{{Third side note: At FUDCon, infrastructure realizes how deeply overcommitted we are. People at "The Next Big Infrastructure Project" were excited about a bunch of cool ideas for new services but we realized that to do any of them, we needed to either get rid of some of our present responsibilities or get new long-term sysadmins to help out.}}
On IRC, post-FUDCon, we talk with glezos and firm up plans a bit more. Someone convinces everyone else that moving a little at a time doesn't help anyone because there's a lot more confusion if we have two separate places to translate at, two procedures to integrate the translations with software, etc. Fedora Infrastructure has a meeting where we figure out how we feel about migrating transifex service to tx.net. We're in favour. glezos to carry the idea for coordination with the l10n team.
At this point things start moving fast. Content is synced to transifex.net for a Proof of Concept, meeting with l10n is setup to talk about migrating. Decision is made to migrate pre-F15 because the FI-hosted transifex is too painful, coordination with developers who will need to pull the translations on newer versions of transifex is talked about, duties are assigned for migrating, setting up teams, packaging the client-side tools for developers, etc. That brings us pretty much to the present.
What I draw from all this is that:
- Although communication was present, it didn't involve everyone touched by
the change. That's pretty hard to achieve in any circumstance but we could try to make it better. From infra's point of view, transifex was being provided by us to the Fedora l10n team which is not entirely accurate. So there needs to be a way to figure out the chain of dependent people.
- In some ways, people were just waiting on a decision. Fedora
Infrastructure was not able to update to the newest transifex because of workflow changes. However, once it was decided that Fedora Infrastructure, therefore, couldn't upgrade (or fix) transifex at all, it was easy and quick for people on l10n to make a decision about changing workflow to get more reliable service.
- It seems that maintainance is harder to recruit people to work on than a
new direction. That new direction can even be a continuation of the current direction (move to a newer transifex on transifex.net) as long as it's seen as a decision needing certain definite actions.
Jared and Paul said in the meeting log multiple times that we needed to avoid people being surprised by the situation, but unfortunately that wasn't very likely. The timing involved made it a certainty that some people were going to be squeezed, surprised, and upset.
I perceive the following items as fait accompli before that meeting happened:
- Fedora Infrastructure decided they could no longer provide translation
services.
- To meet their service obligations, FI put in place a plan to move
translation services to Transifex.net.
- Fedora Project leadership were essentially in agreemen with that plan.
I didn't see the following items as fait accompli before that meeting:
- Future of hosting our own translation tool of any kind on
fedoraproject.org. Clearly we can revisit this decision, clearly people wanted to.
- Ability for resourceful people to step forward and make it possible to
either not move for F15 or move back for F16. For example, you seem to have responded to this by seeking new, on-going resource commitments.
- Commitment of L10n leadership toward any particular plan. People came in
to that meeting with different opinions and agendas, but seemed to come to consensus by the end.
One note here -- in the meeting skvidal made a statement about what would happen if we didn't move to transifex.net which was both harsh and true. But it shows that there's another thing that wasn't quite a fait accompli:
- Infrastructure could have continued to host a transifex instance but we
could not continue to support it (with updates, with fixes for any of the numerous open bugs, with the ability to make it more reliable, etc). skvidal's projection was that such a strategy, while possible, would eventually lead to our instance of transifex preventing work from being done at a time when nothing could be done about it making everybody upset.
Community supported infrastructure doesn't have to be of a lesser service level overall, but things tend to be less formal across the board about change impact on community participants.
One lesson to pull from this situation is for *every* service provider (which inclues L10n, Docs, Infrastructure, et al) to have a change management process in place that:
- Exposes problems (needs for change) early and often. 2. Encourages
discussion and work so decisions (why, how) are visible to all affected. 3. Makes sure affected teams are not surprised by developments.
I'd like to see an infrastructure threat board. Here's an idea of how that would work: The threat board would have services that have no dedicated person working on them as level red, things that have one or two person working on them as level amber, less than five in yellow, and green at five or more. A standing policy would be that things in amber or less are in danger of being dropped and we need to work on how we're going to gracefully drop, outsource, or reassign-drum up help for them (for instance, we can't very well drop koji support so if it was in amber, someone would have to come across from something else that they were doing to work on that). Perhaps once a month, we'd post services that are not green to fedora-devel-announce and if we didn't get its levels to rise in X amount of time, we'd be free to enact our contingency plans if we chose.
-Toshio