Previously users had to enable --log.level=debug to see backtraces. These errors catch everything which can make debugging based on user reports difficult if not impossible.
Enabling --log.level=debug to solve these issues is not useful because users cannot enable it in production for rare errors.
Fixes#8156
use rm_rf to delete dir and shutdown pipeline after run
avoid the use of rescue nil, bad practice
do not mock close as it prevents closing the file an prevents removing it on Windows
cleanup temporary files and relax file name check for Windows
fix paths handling for Windows
flag the read file string as UTF-8 which is what we expect
fix Windows issues
ignore load_average on windows
windows safe URI
cleanup and fix file handling for windows
wait for pipeline shutdown to complete
revert to original puts
cleanups
use environment for windows platform check
fix hash path
wait for pipeline thread to complete after shutdown
* add newlines to generated json
* Implement cloud.id and cloud.auth settings merge to module settings
* Fixes from review plus convert to using Password for any Module Setting
* Review changes
* update modules.asciidoc to include a section on Cloud
* Capitalize Id
* remove unnecessesary require lines
With IntelliJ 2017.02 the gradle output changed from 'build/classes' to 'out/production/classes'. See https://youtrack.jetbrains.com/issue/IDEA-175172 for additional information.
This PR changes the local developement behavior to prefer the classes build with IntelliJ (if they exist) over the classes built directly with gradle.
Fixes#8056
Remove LongGauge and DoubleGauge and Replace with NumberGauge
- This is an unecessary distiction for Gauges and can complicate the serialization. Double/Float should serialize with decimal precision, and Jackson handles this by default.
Add custom witness to match existing Ruby API, such that existing Ruby plugins should not need changes
Remove namespace from LazyDelegatingGauge
- It was only there for logging, but newer versions (not here) will completely remove the concept of namespace.
Part of #7788Fixes#8053
This commit broke the plugin contract with plugins like xpack and caused many issues there. Reverted until it can be better handled.
This reverts commit 92cdbe2dbd.
Fixes#7995
* Add --setup phase for modules
Without the `--setup` flag, the index template for elasticsearch, and the index-pattern, saved searches, visualizations, and dashboards for Kibana will not be published.
fixes#7959
In this patch we unify the IDs reported by LIR with those generated using config_ast.rb
This is a temporary fix until LIR execution is built. It relies on the fact that currently both LIR and config_ast.rb only operate on a single concatenated string of configurations. In the future LIR will compile different files separately, then merge their IRs. For now, however, this solution will hold.
This change also exposes the :id attribute of FilterDelegator to bring it to parity with OutputDelegator and InputDelegator which is required for testing purposes.
Fixes#7930
if force shutdown uses `with_pipelines` then it will never acquire the
`@pipelines_lock` from `shutdown_pipelines` until it terminates.
This PR makes force shutdown use `@pipelines` directly, fixing it.
Fixes#7918
This changes the following:
* DRY Gauge metrics
* Remove Namespace from metrics API (never used, not needed)
* Introduce a 'dirty' flag. This will be used to assist with serialization (in a later commit).
* Change getKey to getName in the API
* Deprecate (only from Java) the future non-supported Guage types.
* Remove NumericGauge and replace with LongGuage and DoubleGuage.
* Bring code coverage back to 100%
Note this change is internally non-passive, however the Java metrics API (recently introduced in 6.0) has not been advertised as a re-usable asset.
Changes in support of #7788Fixes#7872
Fix bug where reloading a pipeline would close the DLQWriter, but
leave the closed version cached in the factory object, stopping
the reloaded pipeline from being able to write to the dead letter
queue.
Fixes#7840
- allow empty config.string
- move all config autocompletion logic to the ConfigStringLoader
- gracefully handle absense of files in path.config
- ensure original_settings are restored in multi_local source after PipelineConfig creation
Fixes#7866
The existing implementation uses the RubyArray as the key for the fast lookup Map. Under the covers the Map implementation is comparing equal operators (many times per event), and JRuby Array equals operator is VERY expensive since it literally walks each value of array for each equality. Based on profiling via YourKit, the JRuby Array equals operator is a very hot method consuming upto 60% of sampled CPU cycles while under high load (and no other CPU dominators).
This change is to use the .hash value of the JRuby Array as the key of the fast lookup Map. This implementation still calls .hash for each and every call, which is also expensive, since it also walks the arrays to compute the hash. However, the equality check of the hash value is very fast, and net gain is significant. Upto a 15% increase of throughput.
Fixes#7772Fixes#7798
