Fix/cluster activate stranded secondary node - #1210

Open

wmousa wants to merge 30 commits into

mainfrom

fix/cluster-activate-stranded-secondary-node

Commits on Aug 17, 2026

fix(cluster-activate): repair stranded nodes instead of failing activation
wmousa
committed
2300add
fix(ha): make node-shutdown FTT capacity check failure-domain aware
wmousa
committed
2574628
fix(cluster-activate): make secondary/tertiary pairing domain-order-independent
wmousa
committed
912eb47
fix(ha): correct failure-domain risk budget for node-shutdown capacity check
wmousa
committed
0b5ad29
fix(cluster-activate): require npcs+2 domains, not npcs+1, at fresh activation
wmousa
committed
b05bcca
fix(node-removal): splice a stranded replica into an existing pairing when no free cross-domain candidate exists
wmousa
committed
f9515a7
fix(node-removal): create-before-destroy in the splice fallback
wmousa
committed
c215e16
fix(node-removal): stop a transient RPC/DNS blip from killing phase 5, and stop masking an incomplete phase 5 on retry

wmousa
and
claude
committed
4c243f2
fix(node-removal): skip already-removed peers with stale jm_ids in phase 5
wmousa
committed
d6bd660
fix(storage_node): disconnect hublvol from secondary and tertiary nodes during teardown

Hamdy-khader
authored and
wmousa
committed
129b12e
fix(node-removal): cascade-vacate a splice target's pre-existing occupant
wmousa
committed
174430e
fix(node-removal): clear the removed node's own stale lvstore_ports
wmousa
committed
acf3ca1
fix(web): return 400, not 500, when node remove/delete is refused
wmousa
committed
43d6ebc
fix(node-removal): also refresh peers that inherit a dead JM via hosting
wmousa
committed
465d683
fix(node-removal): detach evicted peer's hublvol controller, not just its bdev stack
wmousa
committed
43d3a67
fix(node-removal): don't destroy the shared lvstore when only relocating a non-leader replica
wmousa
committed
055125d
fix(web): make SimplyblockCollector inherit prometheus_client's Collector
wmousa
committed
d7603c2
fix(storage_node_ops): update node shutdown logic to exclude pending removal status
Hamdy-khader
committed
1fcbd9d

Commits on Aug 18, 2026

Commits on Aug 19, 2026

Commits on Aug 21, 2026

feat(node-removal): replace the JM name-override hack with SPDK's jc_replace_jm
wmousa
committed
0454142

Navigation Menu

Uh oh!

Fix/cluster activate stranded secondary node - #1210

Fix/cluster activate stranded secondary node#1210
wmousa wants to merge 30 commits into
mainfrom
fix/cluster-activate-stranded-secondary-node

Commits on Aug 17, 2026

fix(cluster-activate): repair stranded nodes instead of failing activation

fix(ha): make node-shutdown FTT capacity check failure-domain aware

fix(cluster-activate): make secondary/tertiary pairing domain-order-independent

fix(ha): correct failure-domain risk budget for node-shutdown capacity check

fix(cluster-activate): require npcs+2 domains, not npcs+1, at fresh activation

fix(node-removal): splice a stranded replica into an existing pairing when no free cross-domain candidate exists

fix(node-removal): create-before-destroy in the splice fallback

fix(node-removal): stop a transient RPC/DNS blip from killing phase 5, and stop masking an incomplete phase 5 on retry

fix(node-removal): skip already-removed peers with stale jm_ids in phase 5

fix(storage_node): disconnect hublvol from secondary and tertiary nodes during teardown

fix(node-removal): cascade-vacate a splice target's pre-existing occupant

fix(node-removal): clear the removed node's own stale lvstore_ports

fix(web): return 400, not 500, when node remove/delete is refused

fix(node-removal): also refresh peers that inherit a dead JM via hosting

fix(node-removal): detach evicted peer's hublvol controller, not just its bdev stack

fix(node-removal): don't destroy the shared lvstore when only relocating a non-leader replica

fix(web): make SimplyblockCollector inherit prometheus_client's Collector

fix(storage_node_ops): update node shutdown logic to exclude pending removal status

Commits on Aug 18, 2026

fix(node-removal): repoint lvol.nodes when a replica relocates, not just the storage-node bookkeeping

fix(node-removal): don't guess a listener port when a non-leader host's lvstore_ports entry hasn't landed yet

fix(node-removal): tear down a vacated peer's per-lvol NVMe-oF subsystems, not just its bdev stack

Commits on Aug 19, 2026

fix(node-removal): record the resolved (override-aware) name in remote_device.jm_bdev

fix(node-removal): retire a JM name override once its consumer rebuilds, not when the replacement host does

fix(node-removal): chain a consumer's inherited JM name through a second removal, not the removed node's own name

fix(node-removal): never pick a replacement JM the consumer already reaches by another path

fix(node-removal): fall back to a colliding JM candidate instead of leaving the slot permanently unfilled

Commits on Aug 21, 2026

feat(node-removal): replace the JM name-override hack with SPDK's jc_replace_jm

Commits on Aug 24, 2026

fix(ha): let a SCHEDULABLE node's queued restart actually restart it

Revert "fix(ha): let a SCHEDULABLE node's queued restart actually restart it"

fix(storage_node_monitor): prevent auto-restart for nodes in SCHEDULABLE status