2024-10-08T08:59:51.707Z

<Jonathan Bailey> Build break occurred yesterday when I merged [#59857](https://github.com/ceph/ceph/pull/59857) yesterday it looks like.

Created a PR to fix this here: <https://github.com/ceph/ceph/pull/60186>
Please can someone give this a review so I can get this merged, thanks.

2024-10-08T09:00:13.618Z

<Jonathan Bailey> Build break occurred yesterday during merging of [#59857](https://github.com/ceph/ceph/pull/59857) yesterday it looks like from overnight email.

Created a PR to fix this here: <https://github.com/ceph/ceph/pull/60186>
Please can someone give this a review so I can get this merged, thanks.

2024-10-08T14:50:45.719Z

<Matthews Jose> hello everyone,
Quick question. When a client reads an object from an OSD in Ceph, how is the session between the client and the OSD managed afterward? Specifically, after the read operation is completed, does the client keep the session open for potential reuse, like for a write to another PG or interacting with other OSDs? How does Ceph decide when to close or reuse these sessions for future operations? Thanks

2024-10-08T14:54:26.750Z

<Casey Bodley> class Objecter has a `std::map<int,OSDSession*> osd_sessions;` for each osd that keeps its connection open. you can search for `close_session()` to see what causes these sessions to close

2024-10-08T15:01:27.746Z

<Matthews Jose> Thank you!

2024-10-08T15:12:56.758Z

<Matthews Jose> @Casey Bodley I checked the function, I understand what triggers the session to close. But my question is more about how frequently does a client close and open a session with an OSD. For example if client0 has a session with OSD1 and it reads obj1 from PG 1.2, can it for example keep the session open to query obj67 in PG1.3 for which the primary is still OSD1. Thank you in advance

2024-10-08T15:19:37.828Z

<Casey Bodley> objecter will reuse the osd connection for requests to different PGs, yeah. it looks like Objecter only calls `close_session()` when the client shuts down or when the OSDMap shows the osd went down. otherwise it won't close/reopen these connections

2024-10-08T15:26:37.766Z

<Matthews Jose> I see. So basically if I used RBD and a VM on top of librados the client would essentially keep a session with all OSDs (most likely) for duration of time, till the VMs shutdown. Doesn't that degrade the performance of the OSD since it has to keep processing session heartbeat messages etc?

2024-10-08T16:07:17.188Z

<Zac Dover> can anyone get to [ceph.io](http://ceph.io)?

2024-10-08T16:07:29.713Z

<Zac Dover> I'm having trouble loading [ceph.io](http://ceph.io).

2024-10-08T16:08:17.458Z

<Ken Carlile> yeah, seems to be either down or slow to respond for me, too

2024-10-08T16:08:57.248Z

<Zac Dover> Thanks, @Ken Carlile.

2024-10-08T16:09:24.469Z

<Casey Bodley> lots of lab services still down after scheduled network maintenance this morning

2024-10-08T16:12:57.089Z

<Eugen Block> Ceph mailing list also doesn’t work for me. Anthony confirmed some connection issues. 

2024-10-08T16:37:22.328Z

<Casey Bodley> that's my understanding, but i may be missing something. i don't know how expensive the heartbeats are, but scaling the number of clients could potentially exhaust osd file descriptor limits