ElasticSearch Support
This provides a plugin to use Jaeger with ElasticSearch. This currently supports ElasticSearch 5.x.
Indices
Indices will be created depending on the spans timestamp. i.e., a span with
a timestamp on 2017/04/21 will be stored in an index named jaeger-2017-04-21
.
ElasticSearch also has no support for TTL, so there exists a script ./es_indices_clean.sh
that deletes older indices automatically. The Elastic Curator
can also be used instead to do a similar job.
Using ./es_indices_clean.sh
Parameters:
- Environment variable TIMEOUT that sets the timeout in seconds for indices deletion (default: 120)
- a number that will delete any indices older than that number in days
- ElasticSearch hostnames
- Example usage:
TIMEOUT=120 ./es_indices_clean.sh 4 localhost:9200
Timestamps
Because ElasticSearch's Date
datatype has only millisecond granularity and Jaeger
requires microsecond granularity, Jaeger spans' StartTime
is saved as a long type.
The conversion is done automatically.
Tags
are nested fields in the
ElasticSearch schema used for Jaeger. This allows for better search capabilities and data retention. However, because
ElasticSearch creates a new document for every nested field, there is currently a limit of 50 nested fields per document.
Shards and Replicas
Number of shards and replicas per index can be specified as parameters to the writer and/or through configs under
./pkg/es/config/config.go
. If not specified, it defaults to ElasticSearch defaults: 5 shards and 1 replica.
This article goes into more information
about choosing how many shards should be chosen for optimization.
Limitations
Separation of Spans and Service:Operation Pairs
The current commit has span
and service:operation
documents under the same index for a given date.
This is to be separated into two indices in the near future in preparation for ElasticSearch v6.0. (#292)
Tag query over multiple spans
This plugin queries against spans. This means that all tags in a query must lie under the same span for a
query to successfully return a trace.
Case-sensitivity
Queries are case-sensitive. For example, if a document with service name ABC
is searched using a query abc
,
the document will not be retrieved.
Testing
To locally test the ElasticSearch storage plugin,
- have ElasticSearch running on port 9200
- run
STORAGE=es make storage-integration-test
in the top folder.
All integration tests also run on pull request via Travis. This integration test is against ElasticSearch v5.4.0.
- The script used in Travis can be found under
./travis/es-integration-test.sh
,
and that script be run from the top folder to integration test ElasticSearch as well.
This script requires Docker to be running.
Adding tests
Integration test framework for storage lie under ../integration
.
Add to ../integration/fixtures/traces/*.json
and ../integration/fixtures/queries.json
to add more
trace cases.