[Fleet] Create API endpoint to set fleet.timeout across all policies

[kpollich]

Ref elastic/elastic-agent#2257
Ref elastic/elastic-agent#2408

In order to support self-managed proxy users who may not be able to control their proxies' timeout configurations, Fleet needs to expose an endpoint that allows them to limit Fleet's long polling timeout to 5 minutes. As part of the above issue/PR, the long polling timeout is being increased to 10 minutes by default for all agents. Users who rely on a proxy with a 5 minute timeout will need a way to revert this change on an opt-in basis.

Create an API + persisted setting, e.g

POST /api/fleet/settings
{
  "agent_check_in_long_poll_interval": "5m|10m" // Only options are 5 or 10 minutes
}

When this setting is updated, all policies need to be redeployed, and each policy generated should include the appropriate fleet.timeout value. See the docs in the above PR for an example:

https://github.com/elastic/elastic-agent/pull/2408/files#diff-f0d296ac94b8457b06ae3376b0642cb39fea46a5c630bcca24701d3464b28b5bR47-R54

[elasticmachine]

Pinging @elastic/fleet (Team:Fleet)

[kpollich]

This can be handled via the Fleet settings framework now, afaik.

[jlind23]

@kpollich shall we change this estimate to an S then?

[nchaulet]

@kpollich @nimarezainia Curious what do you think of storing and eventually adding this as part of the fleet server host model? It seems it's related to the fleet connectivity part

We currently do not have global settings for agent policy, we have per agent policy settings in the agent policy UI or global shared part of the policy fleet server hosts, outputs.

[nimarezainia]

@kpollich @nimarezainia Curious what do you think of storing and eventually adding this as part of the fleet server host model? It seems it's related to the fleet connectivity part

@nchaulet sorry what do you mean by "fleet server host model"? if you are suggesting that this should be configurable as part of the fleet settings in the fleet global settings UI (for each fleet server host defined) - then i tend to agree with that.

[nchaulet]

if you are suggesting that this should be configurable as part of the fleet settings in the fleet global settings UI (for each fleet server host defined) - then i tend to agree with that.

Yes it is what I mean

[nchaulet]

@pchila @cmacknz Does that issue is still relevant I saw elastic/elastic-agent#2408 was closed and there is no support for fleet.timeout in the agent config right?

[cmacknz]

We never ended up changing the checkin timeout to 10 minutes as the descriptions suggests would happen.

Fleet server still defaults to 5 minutes plus some jitter: https://github.com/elastic/fleet-server/blob/7f44587b41293fcd436568700f363c0322949cb7/internal/pkg/config/timeouts.go#L58-L59

Elastic Agent still defaults to 10 minutes expecting Fleet server to return a response before then: https://github.com/elastic/elastic-agent/blob/d354d9f48c92b69aed61fbca5f46853b54cba956/internal/pkg/remote/config.go#L48-L49

Possibly the agent transport settings referenced there can be set in the fleet section of the policy without changing anything in agent, but AFAIK there is no current use case for this besides maybe increasing the number agents fleet can manage at once.

[nchaulet]

Possibly the agent transport settings referenced there can be set in the fleet section of the policy without changing anything in agent, but AFAIK there is no current use case for this besides maybe increasing the number agents fleet can manage at once.

@cmacknz @kpollich should we close this one?

[kpollich]

Yes seems like we should close this. Thanks for flagging.