In T402206#11095526, @Jgiannelos wrote:I think the problem might be the way MW on beta talks to RESTBase. […] Ideally since (other than serving mathoid) RESTBase is not used in production, we should replicate the same behaviour in beta cluster too.
In T402206#11103829, @Krinkle wrote:[…] Two ways come to mind:
- ATS plugin + REST Gateway — This would be most like prod
- in ATS, keep the rb-mw-mangling plugin, which rewrites part of the RESTBase URL already.
- in ATS, add a beta version of the gateway-check.lua plugin to direct various subpaths under /api/rest_v1/ to rest-gateway
- set up REST Gateway somewhere (Helm chart). This may be non-trivial to do in Beta. I suspect it would not actually be abl to reuse much since it seems fairly specific to production, and the indirection would presumably do very little. In prod it takes care of monitoring and in theory might do throttling, but we might not need that in Beta.
- ATS plugin only — This would be similar to prod but simpler.
- in ATS, keep the rb-mw-mangling plugin, which rewrites part of the RESTBase URL already.
- in ATS, expand the rb-mw-mangling-beta plugin to add more rewrites. This would be similar to what the gateway-check.lua plugin and rest-gateway service do together in prod. In other words, instead of rewriting /api/rest_v1/page/summary/(.*) to a rest-gateway call (which then proxies to mobileapps service), rewrite it directly to wherever the mobileapps service runs in beta.
[…]
In T402206#11106759, @Krinkle wrote:I would recommend [the] simpler setup that should achieve the same, with less maintenance, less initial setup, and less risk of breakage. […]
In T402206#11159512, @Krinkle wrote:My idea/patch here is a long-term improvement to solve a pre-existing problem, […]. That is volunteer work and has no timeline. […]
See also: