Browse code
Add whisper-fix-cpu-metrics.
| 1 | 1 |
new file mode 100755 |
| ... | ... |
@@ -0,0 +1,35 @@ |
| 1 |
+#!/bin/bash |
|
| 2 |
+# Fix obviously erroneous CPU metrics in Graphite+collectd Whisper files. |
|
| 3 |
+cd /var/lib/graphite/whisper/collectd/ || exit 50 |
|
| 4 |
+ |
|
| 5 |
+for host in *; do |
|
| 6 |
+ echo "Treating host ${host}:"
|
|
| 7 |
+ cd "${host}" || exit 49
|
|
| 8 |
+ cpu_count=$(find -type d -name 'cpu-*' | wc -l) |
|
| 9 |
+ ((cpu_threshold=110*cpu_count)) |
|
| 10 |
+ echo " ${host} has ${cpu_count} CPUs, considering a threshold of ${cpu_threshold}"
|
|
| 11 |
+ for file in cpu-*/cpu-*.wsp; do |
|
| 12 |
+ echo " Treating ${file}:"
|
|
| 13 |
+ whisper-dump "${file}" | perl -lanE '
|
|
| 14 |
+ # Skip non-data lines: |
|
| 15 |
+ next unless m#^\d+: \d+, [0-9.]+$#; |
|
| 16 |
+ ($timestamp, $value) = ($F[1], $F[2]); |
|
| 17 |
+ # Spot timestamps with abnormal values: |
|
| 18 |
+ if ($value > '${cpu_threshold}') {
|
|
| 19 |
+ $fix_timestamp = $timestamp; |
|
| 20 |
+ next; |
|
| 21 |
+ } |
|
| 22 |
+ # Pick up the next value to replace the abnormal one: |
|
| 23 |
+ if ($fix_timestamp) {
|
|
| 24 |
+ $fix = sprintf(q[%d:%f], $fix_timestamp, $value); |
|
| 25 |
+ push(@fixes, $fix); |
|
| 26 |
+ $fix_timestamp = 0; |
|
| 27 |
+ } |
|
| 28 |
+ END {
|
|
| 29 |
+ if (@fixes) {
|
|
| 30 |
+ printf(qq[ whisper-update '${file}' %s\n], join(q[ ], @fixes));
|
|
| 31 |
+ } |
|
| 32 |
+ }' |
|
| 33 |
+ done |
|
| 34 |
+ cd - > /dev/null || exit 40 |
|
| 35 |
+done |