Apps are missing logs in App Metrics Log Store in Tanzu Application Service for VMs
Article ID: 297943
Updated On:
Products
Issue/Introduction
Symptoms
- Some apps are missing logs in the App Metrics, however there are logs available while checking from the Apps Manager.
- This issue was observed on foundation with Tanzu Application Service (TAS) for VMs v2.9+ and App Metrics: v2.0.5-build.6 or Metric Store: v1.4.4.
How to identify this issue
All of the problem apps have the same partition ID, 0 and 63. For example, issued app GUID: 44f5d477-d3b1-4499-835f-3a59642424cc.vcap@10fddf38-4c12-4fc2-95f4-58f42932ecae:/var/vcap/jobs/log-store/packages/node-locator$ SOURCE_ID=44f5d477-d3b1-4499-835f-3a59642424cc ./node-locator FIELD NAME: TYPE: ENV: REQUIRED: VALUE: Config.Instances map[int]string LOG_STORE_INSTANCES true map[0:10.xxx.xxx.xx:8080 1:10.xxx.xxx.xx:8080 2:10.xxx.xxx.xx:8080] Config.LocalAddr string LOG_STORE_LOCAL_ADDR true 10.xxx.xxx.xx:8080 Config.PartitionCount int PARTITION_COUNT true 64 Config.ReplicationFactor int REPLICATION_FACTOR true 2 Config.SourceID string SOURCE_ID true 44f5d477-d3b1-4499-835f-3a59642424cc Source ID 44f5d477-d3b1-4499-835f-3a59642424cc lives on the following partitions ID ADDRESS 63 10.xxx.xxx.xx:8080 0 10.xxx.xxx.xx:8080
Run the influx_inspect verify command on all log-store VMs, and ensure there was no error message.
Output is similar to the following:
/var/vcap/store/log-store/influxdb/8/data/logs/default/16122240000000/0000003-000001.tsm: healthy Broken Blocks: 0 / 17739, in 0.100052868s
Verify if there is any app data in the given partition table. As expected, there is one partition that doesn't have app data:
vcap@10fddf38-4c12-4fc2-95f4-58f42932ecae:/var/vcap/data/packages/influx-inspect/cfd412d2b0968da3c4031c97c5699e5b3c54e738$ grep -r 44f5d477-d3b1-4499-835f-3a59642424cc /tmp/step5-63 | wc -l 0 vcap@10fddf38-4c12-4fc2-95f4-58f42932ecae:/var/vcap/data/packages/influx-inspect/cfd412d2b0968da3c4031c97c5699e5b3c54e738$ grep -r 44f5d477-d3b1-4499-835f-3a59642424cc /tmp/step5-0 | wc -l 441
Environment
Resolution
Since App Metrics v2 is rearchitected, the log-store is intended to replicate data across the number of nodes per the replication factor configuration. By default it is set to replication factor of 2 across 3 nodes.
In this edge case, logs were not correctly getting written to multiple partitions on a single node. This only happens when a source ID gets hashed onto partition 63 and 0 on log-store-vms/0 in a 3 nodes, 64 partition, 2 replication factor default deployment. This is why we wouldn’t see logs propagated into partition 63.
Workaround
A workaround for this issue would be to change the configuration from 3 nodes to 4 nodes.
Note: Log store does not support horizontal scaling, and this will result in total data loss. We recommend doing this with new disks.
If you require the logs, you can use the downloads endpoint.
# download the gz file to the tmp directory curl -k --insecure --fail --output /tmp/logs.gz -vvv --capath /var/vcap/jobs/log-store/config/certs/ca.crt --cert /var/vcap/jobs/log-store/config/certs/api.crt --key /var/vcap/jobs/log-store/config/certs/api.key https://localhost:8080/v1/sources/<your-source-id>/download # head over to the /tmp folder and unzip cd /tmp gzip -d logs.gz # check logs are uncorrupt and human readable cat logs
Conclusion
For more information, refer to the App Metrics v2.0.6 - Release Notes under Maintenance Updates.
Log Store has the following fixes:
Feedback
