We have upgraded our Performance Management to 21.2.3+ and during the upgrade the database failed to start.
Subsequent attempts to start the db with admintools presents an error message:
..
Database startup failed.
Too many nodes having invalid/missing last good epoch values to continue.
Run diagnostics and contact vertica technical support.
Check vertica.log for these messages:
2022-02-28 11:13:12.301 nameless:7f317db14600 [Txn] <INFO> Looking for catalog at: /catalog/drdata/v_drdata_node0001_catalog/Catalog
2022-02-28 11:13:12.303 nameless:7f317db14600 [Catalog] <INFO> Loading Checkpoint 465619
2022-02-28 11:13:12.304 nameless:7f317db14600 [Catalog] <INFO> From desc file: total 2 = global 1, local 1, shard 0
2022-02-28 11:13:12.304 nameless:7f317db14600 [Catalog] <INFO> From directory: total 2 = global 1, local 1, shard 0
2022-02-28 11:13:12.304 nameless:7f317db14600 [Init] <INFO> Startup [Reading Catalog] Reading Checkpoint (bytes) - 0 / 2438511
2022-02-28 11:13:12.304 nameless:7f317db14600 [Catalog] <INFO> Loading 2 chunks in 2 threads
2022-02-28 11:13:13.034 nameless:7f31733ae700 [Main] <INFO> Handling signal: 4
2022-02-28 11:13:13.172 nameless:7f31733ae700 [Main] <PANIC> Received fatal signal SIGILL.
2022-02-28 11:13:13.172 nameless:7f31733ae700 [Main] <PANIC> Info: si_code: 2, si_pid: 36914407, si_uid: 0, si_addr: 0x2344e7
Run dmesg and look for this error:
traps: bootstrap-catal[331677] trap invalid opcode ip:244431f sp:7ffed0bb4160 error:0 in vertica[400000+814f000]
If you see this, check the architecture of the CPU’s
cat /proc/cpuinfo
check if the CPU type is
AMD Opteron
Release : 21.2.3 and newer
Vertica 10.1.1
Component : Performance Management Upgrades
There is a known defect with vertica where it cannot install on AMD Opteron CPU’s.
Please install on different architecture CPU’s.
An issue has been raised with vertica to update their docset to call this out.
Currently, it states that AMD is supported as a blanket statement: