Config Diff Tool Output

+ added: mon_debug_no_require_umbrella (global.yaml.in)
- removed: mon_debug_no_require_squid (global.yaml.in)

The above configuration changes are found in the PR. Please update the relevant release documentation if necessary.
Ignore this comment if docs are already updated. To make the "Check ceph config changes" CI check pass, please comment /config check ok and re-run the test.

This pull request can no longer be automatically merged: a rebase is needed and changes have to be manually resolved

@markhpc I tagged you to review the RocksDB submodule version

Putting these links here so I don't have to find them again later:

• Tentacle kickoff! #61321
• tentacle kickoff v2 #62089

@bill-scales

Changes so far look good.

src/common/ceph_strings.cc:: ceph_release_features (line 171) will need updating for umbrella because we are changing the client->OSD protocol for both direct reads and pool migration

As in this?

Index: src/common/ceph_strings.cc
IDEA additional info:
Subsystem: com.intellij.openapi.diff.impl.patch.CharsetEP
<+>UTF-8
===================================================================
diff --git a/src/common/ceph_strings.cc b/src/common/ceph_strings.cc
--- a/src/common/ceph_strings.cc	(revision 3fa00db04dda621b52c55aa7dfc883e9659d5494)
+++ b/src/common/ceph_strings.cc	(date 1765271478614)
@@ -166,7 +166,11 @@
 		return req;
 
 	req |= CEPH_FEATUREMASK_CRUSH_MSR;
-	if (r <= CEPH_RELEASE_SQUID)
+	if (r <= CEPH_RELEASE_TENTACLE)
+		return req;
+
+	req |= CEPH_FEATUREMASK_SERVER_UMBRELLA;
+	if (r <= CEPH_RELEASE_UMBRELLA)
 		return req;
 
 	return req;

Do we need new feature bits for DIRECT READS and POOL requests?

jenkins test make check

- [ ] OSDMap::get_min_compat_client: core team evaluate

The client-exposed stuff core might have in Umbrella is EC direct reads and pool migration but these features are pending. The branches are supposed to encapsulate poking with the features bits (hopefully just piggy-backing on SERVER_UMBRELLA). We cannot forget to modify get_min_compat_client() there.

CC: @aainscow.

- [ ] OSDMap::get_min_compat_client: core team evaluate

The client-exposed stuff core might have in Umbrella is EC direct reads and pool migration but these features are pending.

What does "pending" mean?

The branches

You mean PRs against main?

are supposed to encapsulate poking with the features bits (hopefully just piggy-backing on SERVER_UMBRELLA).

I'd like to understand better why this should rely on checking for SERVER_UMBRELLA. Obviously anything in main cannot currently do so.

We cannot forget to modify get_min_compat_client() there.

I don't follow. It sounds like you want to rely on connection features then why do you need to set the min_compat_client?

I'm trying to get more familiar with OSDMap::get_min_compat_client() but I actually can't find in the code where the enforcement occurs (in the OSD?). Can you point it out please?

• OSDMap::get_min_compat_client: core team evaluate

The client-exposed stuff core might have in Umbrella is EC direct reads and pool migration but these features are pending.

What does "pending" mean?

It means the PRs are not merged yet.

The branches
You mean PRs against main?
Yes.

are supposed to encapsulate poking with the features bits (hopefully just piggy-backing on SERVER_UMBRELLA).

I'd like to understand better why this should rely on checking for SERVER_UMBRELLA. Obviously anything in main cannot currently do so.

For EC direct reads, the client doesn't actually look at any feature bits. It looks at a flag in the pool, which can only be set once all OSDs have upgraded to umbrella.

For pool migration, the design is here:
https://docs.ceph.com/en/latest/dev/pool-migration-design/

There is a section on upgrade. This work is not complete, but I do know the developers are aware of the requirements.

Unless there is explicit work to do here, I think we should move this discussion to the pool migration work.

We cannot forget to modify get_min_compat_client() there.

I don't follow. It sounds like you want to rely on connection features then why do you need to set the min_compat_client?

Can we move this discussion to the Pool Migration? I will add you to the necessary slack channel.

I'm trying to get more familiar with OSDMap::get_min_compat_client() but I actually can't find in the code where the enforcement occurs (in the OSD?). Can you point it out please?

I will put you in contact with the necessary.

jenkins test make check

jenkins test dashboard cephadm

This PR is under test in https://tracker.ceph.com/issues/75532.

This pull request can no longer be automatically merged: a rebase is needed and changes have to be manually resolved