diff options
Diffstat (limited to 'health/health.d/hdfs.conf')
| -rw-r--r-- | health/health.d/hdfs.conf | 37 |
1 files changed, 10 insertions, 27 deletions
diff --git a/health/health.d/hdfs.conf b/health/health.d/hdfs.conf index bd8308bed..ca8df31b9 100644 --- a/health/health.d/hdfs.conf +++ b/health/health.d/hdfs.conf @@ -1,28 +1,11 @@ -# make sure hdfs is running - - template: hdfs_last_collected_secs - on: hdfs.heap_memory - class: Storage -component: HDFS - type: Latency - calc: $now - $last_collected_t - units: seconds ago - every: 10s - warn: $this > (($status >= $WARNING) ? ($update_every) : ( 5 * $update_every)) - crit: $this > (($status == $CRITICAL) ? ($update_every) : (60 * $update_every)) - delay: down 5m multiplier 1.5 max 1h - info: number of seconds since the last successful data collection - to: webmaster - - # Common template: hdfs_capacity_usage on: hdfs.capacity - class: Storage + class: Utilization + type: Storage component: HDFS - type: Utilization calc: ($used) * 100 / ($used + $remaining) units: % every: 10s @@ -37,9 +20,9 @@ component: HDFS template: hdfs_missing_blocks on: hdfs.blocks - class: Storage + class: Errors + type: Storage component: HDFS - type: Errors calc: $missing units: missing blocks every: 10s @@ -51,9 +34,9 @@ component: HDFS template: hdfs_stale_nodes on: hdfs.data_nodes - class: Storage + class: Errors + type: Storage component: HDFS - type: Errors calc: $stale units: dead nodes every: 10s @@ -65,9 +48,9 @@ component: HDFS template: hdfs_dead_nodes on: hdfs.data_nodes - class: Storage + class: Errors + type: Storage component: HDFS - type: Errors calc: $dead units: dead nodes every: 10s @@ -81,9 +64,9 @@ component: HDFS template: hdfs_num_failed_volumes on: hdfs.num_failed_volumes - class: Storage + class: Errors + type: Storage component: HDFS - type: Errors calc: $fsds_num_failed_volumes units: failed volumes every: 10s |