Page MenuHomePhabricator

IP address list for grid nodes / Freenode iline request
Closed, ResolvedPublic

Description

On migrating a bot from trusty to stretch it's failing to connect to Freenode with

Closing Link: 185.15.56.1 (Too many user connections (global))

I can submit an iline for this, but since this is shared infra ideally an operator can do it on my behalf.

I believe the exec nodes did have a per-instance IP (at least long ago in history), the stretch cluster appears to use a nat gw or similar as I get the same across exec nodes (didn't dig into this very deeply to confirm however).

This is blocking migration the IRC relay workload off of trusty as external tools rely on the Freenode output.

Please let me know if you require any more details.

Event Timeline

Az1568 triaged this task as High priority.
Az1568 moved this task from Backlog to Group contacts needed on the wikimedia-irc-libera board.
Az1568 subscribed.

This is fundamentally a duplicate of T151704: Libera Chat may throttle bot connections from tools, but the problem is currently worse than it has been in the past because of the lack of public IPv4 addresses on the new Stretch job grid. We should really try T151704#3467490 before we resort to abusing public IPs like we did in the past.

I've contacted freenode to see about resolving the issue in the short term (can't say for certain they'll agree to set the i:line or not), but long term we do need a better solution.... optimally as suggested above we'll need to see about having an ident daemon running so that each individual user with access will have a unique identity for them or their bots.

Hi Alex,

I've added that IP to the list of current Wikimedia IPs, they have a shared
connection limit of 500, I believe that will be enough. If it isn't, poke me
(emerson) on IRC or reply to this ticket.

Thanks,
emerson

Have they added the whole range or just that NAT IP?
If they've just added the NAT IP, I expect problems will resume if the tools exec nodes get given floating IPs?

Have they added the whole range or just that NAT IP?
If they've just added the NAT IP, I expect problems will resume if the tools exec nodes get given floating IPs?

If they've whitelisted our NAT IP, do we still need to give the exec nodes floating IPs? T141445#2600708 suggests that it was for freenode, and our own wikis. I assume we can fix any rate limiting issue on the MediaWiki side (if there still is one?). I would expect all bots to have the noratelimit userright.

Have they added the whole range or just that NAT IP?
If they've just added the NAT IP, I expect problems will resume if the tools exec nodes get given floating IPs?

I believe they just added the NAT IP to the existing i:line for us.

Have they added the whole range or just that NAT IP?
If they've just added the NAT IP, I expect problems will resume if the tools exec nodes get given floating IPs?

If they've whitelisted our NAT IP, do we still need to give the exec nodes floating IPs? T141445#2600708 suggests that it was for freenode, and our own wikis. I assume we can fix any rate limiting issue on the MediaWiki side (if there still is one?). I would expect all bots to have the noratelimit userright.

probably not, let's hope no other use cases for that come up in future?

I think we're all good here?