[8/n] [sync-switch] read switch config in one transaction

[sunshowers]

Part two of the fix for #10640. This wires up what was introduced in the previous commit, #10648, dropping the piecemeal reads. We aren't totally out of the woods yet, because the transactional read currently doesn't handle transient db issues as well as it could. We're going to fix that in a principled fashion in upcoming commits.

Note that I haven't touched the mgd and other parts of the switch config in this commit, since @jgallagher is actively working on them. (Let me know if I should also include those changes either here or in a followup.)

Depends on:

• [1/n] [sync-switch] build bootstore bgp peers from already-loaded port settings #10642
• [2/n] [sync-switch] move bootstore config building into nexus-switch-config #10643
• [3/n] [sync-switch] move bfd and address lot reads back to bg task, make the builder pure #10644
• [4/n] [sync-switch] replace db-model builder inputs with domain types #10645
• [5/n] [sync-switch] extract db translation into a new crate #10646
• [6/n] [sync-switch] factor switch_port_settings_get body out #10647
• [7/n] [sync-switch] add single-transaction switch config read (not yet wired up) #10648

[jgallagher]

Note that I haven't touched the mgd and other parts of the switch config in this commit, since @jgallagher is actively working on them. (Let me know if I should also include those changes either here or in a followup.)

The work I'm doing can be summarized as "move any progenitor/HTTP interactions between nexus and {mgd,dpd} out of sync_switch_configuration and into sled-agent-scrimlet-reconcilers". I'm mostly not touching the types (although I'm filing some issues as I run into problems) nor anything that involves the db - if there are similar db/transaction-safety issues with other parts of the switch config, I'd vote for fixing those too. The scrimlet reconcilers project is mostly only a consumer of the bootstore; it expects Nexus to construct a correct one.

[jgallagher]

Note that I haven't touched the mgd and other parts of the switch config in this commit, since @jgallagher is actively working on them. (Let me know if I should also include those changes either here or in a followup.)

The work I'm doing can be summarized as "move any progenitor/HTTP interactions between nexus and {mgd,dpd} out of sync_switch_configuration and into sled-agent-scrimlet-reconcilers". I'm mostly not touching the types (although I'm filing some issues as I run into problems) nor anything that involves the db - if there are similar db/transaction-safety issues with other parts of the switch config, I'd vote for fixing those too. The scrimlet reconcilers project is mostly only a consumer of the bootstore; it expects Nexus to construct a correct one.

Ohhh I might have misunderstood. Is this about this comment?

omicron/nexus/src/app/background/tasks/sync_switch_configuration.rs

Lines 616 to 621 in 4718e46

	// NOTE: this mgd-apply path re-reads each peer's
	// communities and import/export policies via
	// `communities_for_peer` / `allow_*_for_peer`, even
	// though `switch_port_settings_get` already loaded
	// them. That redundancy could be removed too, but is
	// left for now.

and the related code? If so, yes, you can ignore this for now - all of that is definitely under the "progenitor interaction between nexus and mgd" umbrella. It does use the db but not to construct the bootstore, and yeah we'll be able to just delete it all once the scrimlets are responsible for mgd config sync'ing.

[sunshowers]

and the related code? If so, yes, you can ignore this for now - all of that is definitely under the "progenitor interaction between nexus and mgd" umbrella. It does use the db but not to construct the bootstore, and yeah we'll be able to just delete it all once the scrimlets are responsible for mgd config sync'ing.

Yeah, exactly -- that's what I figured.