Server-native maintenance notifications: opt in, react, hand off, and recover when nothing is announced - #3191

Open

mgravell wants to merge 42 commits into

mainfrom

marc/maint-optin-server

Commits on Aug 25, 2026

Toy server: support the maintenance-notification opt-in, and sending events
mgravell
committed
0f3ae13

Commits on Sep 1, 2026

Maintenance notifications: opt in, and receive (D2) (#3193)
mgravell
authored
8109d4b
D8: document the notifications, and make the diagnostic it recommends work (#3204)
mgravell
authored
3d9b3a6
Ask for a moving-endpoint-type, and find the successor field was never empty
mgravell
committed
b44132a
Derive moving-endpoint-type per connection
mgravell
committed
901f067
Use a named successor directly, instead of hoping DNS agrees
mgravell
committed
fb7160a
Default to deriving the moving-endpoint-type
mgravell
committed
7d8f805
Cover mTLS, which also proves the TLS half of the derivation
mgravell
committed
c007915

Commits on Sep 2, 2026

Apply the half-window reconnect rule where it is the only tool
mgravell
committed
8f6577f
Serialise the fault-injector tier, and stop bounding it with an HTTP timeout
mgravell
committed
22ebea4
Sweep the databases a killed scenario leaves behind
mgravell
committed
b5e6061
D9: a soak for the state that only misbehaves with repetition
mgravell
committed
e9830ac
Prove relaxed timeouts rescue a command, and attribute one that fails
mgravell
committed
c733b1c
Measure how long a completion stays retained
mgravell
committed
3e975df
Key the handoff count on distinct MOVING sequences, not one overall
mgravell
committed
d033d9c
A catch-up completion opens no relaxed window
mgravell
committed
e3ad81c
Read the retention probe from the log, not the relaxed window
mgravell
committed
75fa3e5
Stop claiming parity with go-redis on the post-event tail
mgravell
committed
3572a70
Say that the connect-time rule is completions only
mgravell
committed
51d22f0
Retention measured to three hours, not ninety minutes
mgravell
committed
acd8309

Commits on Sep 3, 2026

Attribute a fault to a window that has just closed, and warn on a missed handoff
mgravell
committed
8a38780
Do not raise an event for a replayed completion
mgravell
committed
c246f9f
Run the destructive scenarios, supervised
mgravell
committed
7cc75fe
Document the destructive gate in the tier README
mgravell
committed
b1aea16
Prove a refusing node is retired, and correct why it was in doubt
mgravell
committed
366c9fe
Put the fault-injector tier in the solution
mgravell
committed
13bb7d3
Re-read the topology when an endpoint only ever refuses
mgravell
committed
5d603ae
One black-hole tunnel, not one per test class
mgravell
committed
f0de7f4
Document the connect-failure topology refresh
mgravell
committed
956ce53
Merge main: keep both sets of unshipped public API entries
mgravell
committed
31c96d0
Docs: lead with the server-native notifications, and name the feature
mgravell
committed
fa6b032
Soften the SequenceId wording
mgravell
committed
eb8c4a0
Validate endpoint retirement against a real node removal
mgravell
committed
a9d3f1c
cluster_failure ends the deployment; assert clean degradation, not recovery
mgravell
committed
f7b1f97
Warn that a stale environment directory passes every check
mgravell
committed
1a4c858

Commits on Sep 8, 2026

Re-read the topology on a long timer as a backstop
mgravell
committed
16adfe3
Say what ConfigCheckSeconds actually checks
mgravell
committed
377abfa

Commits on Sep 11, 2026

Merge main (3.2.0 shipped): keep only what is genuinely unshipped
mgravell
committed
e0c0b0c