We run a two-node Manticore cluster, ar_cluster, with 31 tables and roughly 6 TiB of index data. All application writes go to archive01. Archive02 is intended to be a near-real-time backup. Both nodes were reporting synced before planned maintenance.
After a clean shutdown and restart, archive02 logged no nodes found for ar_cluster, attempted to bootstrap itself, and was rejected by Galera’s bootstrap safety check. We then stopped both nodes, restored the cluster entries in manticore.json, started archive01 as the bootstrap node, and started archive02 normally. Archive02 joined as a snapshot receiver with last_committed = -1.
The first SST reached mail_2024, then failed around November 11, 2026, 12:40 a.m. Toronto time. Archive02 logged retries limit exceeded and not JOINING when sst_received() called, state: CLOSED. Archive01 logged that the receiver left the group and the transfer was canceled.
After a normal restart of archive02, a fresh SST began. Archive02’s searchd then aborted at 1:15:00 a.m. Toronto time (PID 668140, SIGABRT, signal 6). The stack includes malloc() → SstProgress_c::UpdateFromRemote → ReceiveSstProgress. Manticore’s watchdog restarted the process. The replacement crashed again at 1:36:05 a.m. (PID 670715, SIGSEGV, signal 11). Both crash dumps identify an SST progress request for mail_2022. We did not find a corresponding OOM kill or disk I/O error in the kernel logs checked.
The server build is 17.5.1 d4cc0969e@26020616, with Columnar and Secondary 13.0.0 and Galera 3.37. Archive01 is now primary / synced, cluster size 1, with last_committed = 155179395 at the last check. Archive02 is stopped and its grastate.dat is zero bytes. We have not edited it.
Can you identify the SST crash and confirm a fixed, compatible build and upgrade sequence? Is there a supported way to seed archive02 from a consistent copy of archive01 while preserving the correct replication position? Separately, can you advise why archive02’s saved peer information was missing on the first restart and how to verify that it persists before future maintenance?