diff options
Diffstat (limited to 'conf.d/health.d/memory.conf')
| -rw-r--r-- | conf.d/health.d/memory.conf | 30 |
1 files changed, 30 insertions, 0 deletions
diff --git a/conf.d/health.d/memory.conf b/conf.d/health.d/memory.conf new file mode 100644 index 000000000..3c904f6b1 --- /dev/null +++ b/conf.d/health.d/memory.conf @@ -0,0 +1,30 @@ + + alarm: 1hour_ecc_memory_correctable + on: mem.ecc_ce + lookup: sum -10m unaligned + units: errors + every: 1m + warn: $this > 0 + delay: down 1h multiplier 1.5 max 1h + info: number of ECC correctable errors during the last hour + to: sysadmin + + alarm: 1hour_ecc_memory_uncorrectable + on: mem.ecc_ue + lookup: sum -10m unaligned + units: errors + every: 1m + crit: $this > 0 + delay: down 1h multiplier 1.5 max 1h + info: number of ECC uncorrectable errors during the last hour + to: sysadmin + + alarm: 1hour_memory_hw_corrupted + on: mem.hwcorrupt + calc: $HardwareCorrupted + units: MB + every: 10s + warn: $this > 0 + delay: down 1h multiplier 1.5 max 1h + info: amount of memory corrupted due to a hardware failure + to: sysadmin |