Merge remote-tracking branch 'origin/bugfix/S3UTILS-150-fix-bucket-ca…

…pacity-alerts' into w/1.14/bugfix/S3UTILS-150-fix-bucket-capacity-alerts
scality · Nov 28, 2023 · 81386bc · 81386bc
2 parents 3f11858 + 7d2df0b
commit 81386bc
Show file tree

Hide file tree

Showing 5 changed files with 10 additions and 56 deletions.
diff --git a/monitoring/update-bucket-capacity-info-cronjob/alerts.test.yaml b/monitoring/update-bucket-capacity-info-cronjob/alerts.test.yaml
@@ -3,33 +3,6 @@ rule_files:
   - alerts.rendered.yaml
 
 tests:
-  - name: Update Bucket Capacity Info CronJob Test Taking Too Long
-    interval: 1m
-    input_series:
-      - series: 'kube_job_status_start_time{job="kube-state-metrics", job_name="artesca-data-ops-update-bucket-capacity-info-1", namespace="zenko"}'
-        values: '0x9'
-      - series: 'kube_job_status_completion_time{job="kube-state-metrics", job_name="artesca-data-ops-update-bucket-capacity-info-1", namespace="zenko"}'
-        values: '_x4 240x5'
-      - series: 'kube_job_status_start_time{job="kube-state-metrics", job_name="artesca-data-ops-update-bucket-capacity-info-2", namespace="zenko"}'
-        values: '_x4 240x5'
-      - series: 'kube_job_status_completion_time{job="kube-state-metrics", job_name="artesca-data-ops-update-bucket-capacity-info-2", namespace="zenko"}'
-        values: '_x9 '
-    alert_rule_test:
-      - alertname: UpdateBucketCapacityJobTakingTooLong
-        eval_time: 4m
-        exp_alerts: []
-      - alertname: UpdateBucketCapacityJobTakingTooLong
-        eval_time: 9m
-        exp_alerts:
-          - exp_labels:
-              severity: warning
-              job_name: artesca-data-ops-update-bucket-capacity-info-1
-            exp_annotations:
-              description: |
-                Job artesca-data-ops-update-bucket-capacity-info is taking more than 240s to complete.
-                This may cause bucket capacity to be out of date and Veeam SOSAPI avalability as risk.
-              summary: update-bucket-capacity-info cronjob takes too long to finish
-
   - name: Update Bucket Capacity Info CronJob Test No Success in 10m
     interval: 1m
     input_series:

diff --git a/monitoring/update-bucket-capacity-info-cronjob/alerts.yaml b/monitoring/update-bucket-capacity-info-cronjob/alerts.yaml
@@ -16,21 +16,6 @@ x-inputs:
 groups:
   - name: update-bucket-capacity-info-cronjob/alerts.rules
     rules:
-      - alert: UpdateBucketCapacityJobTakingTooLong
-        expr: |
-          time() -
-          (sum by(job_name) (kube_job_status_failed{job_name=~"${update_bucket_capacity_info_cronjob}.*"})
-          > sum by(job_name) (kube_job_status_completion_time{job_name=~"${update_bucket_capacity_info_cronjob}.*"})
-          or sum by(job_name) (kube_job_status_completion_time{job_name=~"${update_bucket_capacity_info_cronjob}.*"}))
-          > ${update_bucket_capacity_info_job_duration_threshold}
-        labels:
-          severity: warning
-        annotations:
-          description: |
-            Job ${update_bucket_capacity_info_cronjob} is taking more than ${update_bucket_capacity_info_job_duration_threshold}s to complete.
-            This may cause bucket capacity to be out of date and Veeam SOSAPI avalability as risk.
-          summary: update-bucket-capacity-info cronjob takes too long to finish
-
       - alert: NoSuccessfulUpdateBucketCapacityJobRunIn10m
         expr: |
           time() 

diff --git a/tests/unit/CompareRaftMembers/BucketStream.js b/tests/unit/CompareRaftMembers/BucketStream.js
@@ -4,7 +4,6 @@ const { versioning } = require('arsenal');
 
 const BucketStream = require('../../../CompareRaftMembers/BucketStream');
 
-const HTTP_TEST_PORT = 9090;
 const TEST_BUCKET_FILTERED_CONTENTS_LENGTH = 3210;
 const DUMMY_VERSION_ID = '987654321 FOOBAR42.42';
 const VERSIONED_RANGE = [1210, 1310];
@@ -183,7 +182,7 @@ describe('BucketStream', () => {
             }
             throw new Error(`unexpected request path ${url.pathname}`);
         });
-        httpServer.listen(HTTP_TEST_PORT);
+        httpServer.listen(0);
     });
     afterAll(done => {
         httpServer.close(done);
@@ -230,7 +229,7 @@ describe('BucketStream', () => {
         test(testCase.desc, done => {
             const bucketStream = new BucketStream({
                 bucketdHost: 'localhost',
-                bucketdPort: HTTP_TEST_PORT,
+                bucketdPort: httpServer.address().port,
                 bucketName: 'test-bucket',
                 marker: testCase.marker,
                 lastKey: testCase.lastKey,
@@ -274,7 +273,7 @@ describe('BucketStream', () => {
     test('listing should continue when all keys in a page are ignored', done => {
         const bucketStream = new BucketStream({
             bucketdHost: 'localhost',
-            bucketdPort: HTTP_TEST_PORT,
+            bucketdPort: httpServer.address().port,
             bucketName: 'bucket-with-replay-keys',
             retryDelayMs: 50,
             maxRetryDelayMs: 1000,

diff --git a/tests/unit/CompareRaftMembers/DiffStreamOplogFilter.js b/tests/unit/CompareRaftMembers/DiffStreamOplogFilter.js
@@ -5,8 +5,6 @@ const { shuffle } = require('arsenal');
 
 const DiffStreamOplogFilter = require('../../../CompareRaftMembers/DiffStreamOplogFilter');
 
-const HTTP_TEST_PORT = 9090;
-
 class MockRaftOplogStream extends stream.Readable {
     constructor(entriesToEmit, refreshPeriodMs) {
         super({ objectMode: true });
@@ -71,7 +69,7 @@ describe('DiffStreamOplogFilter', () => {
             }
             throw new Error(`unexpected request path ${url.pathname}`);
         });
-        httpServer.listen(HTTP_TEST_PORT);
+        httpServer.listen(0);
         httpServer.once('listening', done);
     });
     afterAll(done => {
@@ -81,7 +79,7 @@ describe('DiffStreamOplogFilter', () => {
     test('filtering test with mocks', done => {
         const oplogFilter = new DiffStreamOplogFilter({
             bucketdHost: 'localhost',
-            bucketdPort: HTTP_TEST_PORT,
+            bucketdPort: httpServer.address().port,
             maxBufferedEntries: 5,
             excludeFromCseqs: {
                 1: 10,
@@ -204,7 +202,7 @@ describe('DiffStreamOplogFilter', () => {
     test('should handle an empty input stream', done => {
         const oplogFilter = new DiffStreamOplogFilter({
             bucketdHost: 'localhost',
-            bucketdPort: HTTP_TEST_PORT,
+            bucketdPort: httpServer.address().port,
             maxBufferedEntries: 5,
             excludeFromCseqs: {},
         });
@@ -224,7 +222,7 @@ describe('DiffStreamOplogFilter', () => {
     test('should emit a stream error if failing to fetch RSID after retries', done => {
         const oplogFilter = new DiffStreamOplogFilter({
             bucketdHost: 'localhost',
-            bucketdPort: HTTP_TEST_PORT,
+            bucketdPort: httpServer.address().port,
             maxBufferedEntries: 5,
             excludeFromCseqs: {
                 1: 10,

diff --git a/tests/unit/CompareRaftMembers/RaftOplogStream.js b/tests/unit/CompareRaftMembers/RaftOplogStream.js
@@ -2,7 +2,6 @@ const http = require('http');
 
 const RaftOplogStream = require('../../../CompareRaftMembers/RaftOplogStream');
 
-const HTTP_TEST_PORT = 9090;
 const TEST_OPLOG_NB_RECORDS = 123;
 const DUMMY_VERSION_ID = '987654321 FOOBAR42.42';
 
@@ -82,7 +81,7 @@ describe('RaftOplogStream', () => {
             }
             throw new Error(`unexpected request path ${url.pathname}`);
         });
-        httpServer.listen(HTTP_TEST_PORT);
+        httpServer.listen(0);
         httpServer.on('listening', done);
     });
     afterAll(done => {
@@ -108,7 +107,7 @@ describe('RaftOplogStream', () => {
         test(testCase.desc, done => {
             const oplogStream = new RaftOplogStream({
                 bucketdHost: 'localhost',
-                bucketdPort: HTTP_TEST_PORT,
+                bucketdPort: httpServer.address().port,
                 raftSessionId: 1,
                 startSeq: testCase.startSeq,
                 refreshPeriodMs: 100,
@@ -163,7 +162,7 @@ describe('RaftOplogStream', () => {
         const oplogStream = new RaftOplogStream({
             bucketdHost: 'localhost',
             // change port to get connection errors
-            bucketdPort: HTTP_TEST_PORT + 1,
+            bucketdPort: httpServer.address().port + 1,
             raftSessionId: 1,
             startSeq: 42,
             retryDelayMs: 10,