-
Notifications
You must be signed in to change notification settings - Fork 654
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
Notifying instead of failing fixes #2111 #2112
Notifying instead of failing fixes #2111 #2112
Conversation
Frontend tests were OK 👍 (details) |
Frontend tests were OK 👍 (details) |
+1 and I wonder when we'll fully delete that migrator20.rb and migrator21.rb... probably with 3.0 |
The only user of this is/was the tiler, so probably @rochoa can confirm the Redis keys are not used anymore |
I'm actually removing old api (/tiles/:table/...), check: CartoDB/Windshaft-cartodb#257 That is using the privacy metadata but since it was already deprecated I don't care too much about those endpoints failing randomly due to that change. |
@rochoa : I think we should test this in staging with QA people then so nothing too harmful arises, could you tell me what should we test, related to the tiler? Thanks :-) |
If table privacy changes from "public" to "private" it will fail to render the tile either because the privacy is not enough (from the redis metadata) or because it has not enough database privileges. I will double check the different paths and will write here what to test. |
Ok, thank you very much. I'll wait for this to continue. I'm sorry for asking, but could you give an ETA? This Redis error is also raised at common data metadata loading. I was waiting to end this task to enable metadata loading, but if it's taking some days I might patch it instead. |
What's that "Redis error is also raised at common data metadata loading" error? I'm not aware of it. ETA: I will do it tomorrow. 😄 |
I'm speaking about #2134, I should've linked it before since I spoke about it, sorry 😅 |
hey, if the old api fails is not a problem |
…il_because_of_key_on_table_rename
Frontend tests were OK 👍 (details) |
@rochoa : @Xatpy found a problem in staging (you can check now at ded01) that I've confirmed in local. This branch makes Varnish invalidation stop working, at least at one case. I've checked the changes and it doesn't seem to be a straightforward problem (I mean, I haven't deleted any call to Varnish invalidation), so you can probably help me or, at least, point to me the place where that cache should be invalidated. Steps to reproduce:
Thank you in advance :) PD: I'm trying to reproduce it again in local but I just can't. Nevertheless ded01 won't work. |
raise 'table privacy cannot be nil' unless privacy | ||
# TODO: Improve this, hack because tiler checks it | ||
$tables_metadata.hset redis_key, 'privacy', privacy_for_redis |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
BTW I would remove the privacy_for_redis
method
@juanignaciosl let me look at the Varnish issue, it could have been my fault |
Things I've been testing: Preconditions:
Test scenario:
Although we are not HTH. |
What I mean is that the |
Have a look at this: #1629 |
It is similar, but it won't appear, no matter how much I wait. |
If data changed the {layergroupid}:{last_updated_at} tuple should change so it should not hit anything cached. Are you using private tables? |
I created a new table at both with the defaults for my users, public in staging and private in local, but in local a public one also works. After drawing the polygon I see a difference within two POST requests. This request:
but the similar one in local (
In the staging request there's cdn information, and |
For future reference and the rest of you: @rochoa has been helping me with this (thanks!). The problem was ded01 was using a Windshaft version from June. After upgrading it works as expected. Will continue testing then. |
Frontend tests were OK 👍 (details) |
Yes, mystery solved! As @juanignaciosl said: Maps API in ded01 machine was running an old version with no x-cache-channel header in layergroups. Thanks to @zenitraM all staging servers are being deployed, previously only api01-st was being deployed. So I hope no more issues like this inconsistency in staging environment! |
Got +1 from QA, I'm merging & deploying. |
…n_table_rename Notifying instead of failing fixes #2111
@rambo what do you think about this? I think you told me Redis metadata (behind
$table_metadata
) is no longer used. There're situations like the reported bug when key (table name) is missing and that caused to fail. I think recovering here is legit. I'm not confident about fully removing Redis table metadata usage yet, that's a whole different story now.What to test: registration was previously done at Redis, and that's been removed. Everything should work fine. With an eye on Rollbar to avoid adding new errors.