We have outdated static versions of api/exchanges references here: * http://schemas.taskcluster.net/base/v1/api-reference.json * http://schemas.taskcluster.net/base/v1/exchanges-reference.json They are both out-of-sync with our current in-tree versions: * https://github.com/taskcluster/taskcluster-lib-api/blob/4261a47109e24fdfaec64b0e8fc2578de1e1eb70/src/schemas/api-reference.json * https://github.com/taskcluster/pulse-publisher/blob/bc34bf41bfb5920b0eeb8d6e7a98af325e3280b7/src/schemas/exchanges-reference.json We should make sure that when the versions in github are updated on the master branch of their respective repositories, that the s3 bucket is updated with the latest version of the updated schema. At this time, we may also wish to consider changing the source formal to yml, to be consistent with all the other schemas we store under source control.

Best now to figure out how to solve this in a redeployable fashion. Currently the plan is to just not publish schemas, pending a better design.

I'm going to manually copy across for now, to fix the errors in client generation, but keeping the bug open for the fix in automation.

Updating the schemas uncovered a bug in gojsonschema... https://github.com/xeipuuv/gojsonschema/issues/200 :'-(

Good news - I found an open PR that seems to fix the issue for me - so hopefully that should get merged at some point. https://github.com/xeipuuv/gojsonschema/pull/196

I don't consider this a duplicate, because that bug was fixed, yet the problem persists. Today I manually updated the s3 copies of these files, because our production references still point to http://schemas.taskcluster.net/base/v1/api-reference.json If we are no longer using http://schemas.taskcluster.net/ and http://references.taskcluster.net/ we should delete the buckets rather than leave stale content in there. Dustin, can you confirm if current production service deployments are still publishing documents to http://schemas.taskcluster.net/ and http://references.taskcluster.net/ - or if these are just stale files being served that should be deleted?

Services running for taskcluster.net still deploy to those buckets, yes. However, the "api-reference.json" file was always uploaded by hand -- there's no automation for that. So I think what you did was exactly the right fix.

(In reply to Dustin J. Mitchell [:dustin] pronoun: he from comment #8) > Services running for taskcluster.net still deploy to those buckets, yes. > However, the "api-reference.json" file was always uploaded by hand -- > there's no automation for that. That is what this bug is about, it should be automated! :-)

Sorry, my bad - I was thinking this was about *all* reference files - of course it was just about the single api-references.json file which rarely changes, and we are deprecating the old s3 publish with redeployability, so we can probably mark this as P5 (just in case we never *do* redeployability due to other company priorities).

Again this bit us in bug 1491325 - so I'm moving this back to regular priority - it keeps happening that updates are made, and never get rolled out.

https://github.com/taskcluster/taskcluster-lib-api/pull/120 If that's good, we can do a similar thing for pulse-publisher.

RFC128 changed the plan here -- these are now deployed from tc-lib-references.