Closed Bug 907610 Opened 12 years ago Closed 12 years ago

[bunker] Bunker is not transporting logs to the ES indexer nodes

Categories

(Infrastructure & Operations :: IT-Managed Tools, task, P3)

x86_64
Linux

Tracking

(Not tracked)

RESOLVED FIXED

People

(Reporter: ezounes, Assigned: dmaher)

Details

After a day of debugging this issue, I think I may be close to figuring this out. Here are the symptoms: * Logs are shipped to logstash and structured. They are not send to ES indexer nodes. * If logstash is restarted, logs will be sent for a brief period of time. (Roughly 5 seconds) * I see the following discovery error in the indexer logs: [2013-08-21 07:12:14,493][WARN ][discovery.zen.ping.multicast] [node55_scl3] failed to receive confirmation on sent ping response to [[Fetter, Philip][lRrmOKpmQOipV0EQ_mO2BA][inet[/10.22.78.113:9300]]{client=true, data=false}] Current conclusion: * For some reason the logstash ES clients start out with the correct indexer information, but the discovery protocol fails for some reason. I don't know what changed in ES between 2013/08/20 and 2013/08/21, but logs stopped being indexed at around the same time i pushed an update to the indexers. I've reverted all the changes such that all configs are in their original state before this issue.
Assignee: server-ops-webops → dmaher
Priority: -- → P3
Ugh. I really don't know what the deal is. Perhaps now would be a good time to just go ahead and re-start the interface nodes (or, indeed, the entire cluster).
I understand that this has been fixed ? Could you please provide a quick summary of the analysis and solution ? Thank you.
Flags: needinfo?(ezounes)
I'm fairly certain that this issue was actually caused by removing fields from the default mapping. Here is the sequence of events that might have led to it breaking: 1. The @type field was removed instead of @fields.type 2. The previous mapping dictates that @type should exist. Fields can't be removed from a mapping. 3. Logs stopped being indexed. A bug in Logstash may have caused the logs to be indexed for a brief period of time due to the "drop @type" filter not being applied immediately. Note: Document fields can't be dropped once they're added. However, fields can be added at any time (given dynamic mapping) Thus, if a change is made which changes the structure of a document by removing a field, it will break the mapping. Solution: This issue was resolved by upgrading the ES cluster. I experienced an issue with the same symptoms on 09/03. I resolved it by removing the index. Any time we could make a change that involves modifying an existing type mapping we should wait until we have a fresh index to reduce the amount of log data we lose.
Status: NEW → RESOLVED
Closed: 12 years ago
Flags: needinfo?(ezounes)
Resolution: --- → FIXED
You need to log in before you can comment on or make changes to this bug.