- Update to final v2.0.0 release (bsc#1067341):
+ Compared to version 1.x, this release includes a completely rewritten
storage engine, huge performance improvements, but also many backwards
incompatible changes. For more information, read the announcement blog
post and migration guide:
- https://prometheus.io/blog/2017/11/08/announcing-prometheus-2-0/
- https://prometheus.io/docs/prometheus/2.0/migration/
+ Changes:
* Completely rewritten storage layer, with WAL. This is not backwards
compatible with 1.x storage, and many flags have changed/disappeared.
* New staleness behavior. Series now marked stale after target scrapes
no longer return them, and soon after targets disappear from service
discovery.
* Rules files use YAML syntax now. Conversion tool added to promtool.
* Removed count_scalar, drop_common_labels functions and keep_common
modifier from PromQL.
* Rewritten exposition format parser with much higher performance. The
Protobuf exposition format is no longer supported.
* Example console templates updated for new storage and metrics names.
Examples other than node exporter and Prometheus removed.
* Admin and lifecycle APIs now disabled by default, can be reenabled via
flags
* Flags switched to using Kingpin, all flags are now --flagname rather
than -flagname.
+ Features:
* Remote read can be configured to not read data which is available
locally. This is enabled by default.
* Rules can be grouped now. Rules within a rule group are executed
sequentially.
* Added experimental GRPC apis
* Add timestamp() function to PromQL.
+ Enhancements:
* Remove remote read from the query path if no remote storage is
configured.
* Bump Consul HTTP client timeout to not match the Consul SD watch
timeout.
* Go-conntrack added to provide HTTP connection metrics.
+ Bug Fixes:
* Fix connection leak in Consul SD.
- Update to v2.0.0-rc.3 (bsc#1067030):
+ Enhancements:
* Remove remote read from the query path if no remote storage
is configured.
* Bump Consul HTTP client timeout to not match the Consul SD
watch timeout.
* Bump up a too small max block duration to the min block
duration instead of returning an error.
+ Bug Fixes:
* Avoid needless padding of 4 zero bytes in index files.
* Delete old blocks during reload (necessary on MS Windows).
* Fix regression of alert rules state loss on config reload.
* Serialize background WAL operations to avoid data races.
+ This also contains bugfixes and remote-storage features from
the 1.8 branch.
- Update to v2.0.0-rc.2 (bsc#1065377):
+ Enhancements:
* Handle WAL segments with corrupted header gracefully
* Stabilize memory usage during WAL replay
+ Changes:
* Prefix all storage metrics with prometheus_
+ Bug Fixes:
* Correctly handle label removal in remote read
* Fix chunk misalignment causing out-of-order samples
* Fix connection leak in Consul SD
* Handle invalid chunk dereferences gracefully
* Prevent potential deadlock during failing querier construction
- Update to v2.0.0-rc.1:
+ Features/Enhancements:
* Added a warning for time-drift between the browser and the
prometheus-server.
* Much faster WAL read-back on restart.
+ Bug Fixes:
* Fixed Remote-read to not drop the first series.
* Validate recording-rule names.
* Fix several races.
* Only close blocks if there are no iterators accessing it.
- Refresh 0002-Default-settings.patch
- Update to v2.0.0-rc.0:
+ Numerous changes to the new storage layer, the main changes being:
* Remove `count_scalar`, `keep_common` and `drop_common_labels` functions
* Breaking change in the index format for better consistency
* Fix panic due garbage collected mmap'ed strings
* Fix broken snapshots and admin APIs
* Send HTTP Accept header when scraping
* Use the WAL flush interval passed instead of the hardcoded value
+ This release requires a clean storage directory and is not compatible
with files created by previous beta releases
- Refresh 0002-Default-settings.patch
- Fix loopback address for REST API gateway (bsc#1059462)
+ Added pr-3174.patch
- Update to v2.0.0-beta.5:
+ Bug Fixes:
* Remove deadlock on startup when restoring WAL
* Fix semantical races resulting in invalid persisted files
* Correctly read back WAL in certain edge cases
* Prevent crashes caused by changing metric representations in target's
/metrics
+ Enhancements:
* Overall memory usage reduction
* Serve debugging endpoints while TSDB is loading
* Healthy endpoint correctly reflects liveness during startup
* Switch to consistent usage of go-kit/log
+ This release may have issues with files written by previous beta releases
- Refresh 0002-Default-settings.patch to apply cleanly again
- Update to v2.0.0-beta.4:
+ Numerous changes to the new storage layer, the main changes being:
* Single, compacted write ahead log
* Single in-memory block with garbage collection
* Improve compression of index files
* Cache series dropped via metric_relabel_configs
* Pool byte buffers for scraping
+ Overall the changes achieve a baseline reduction in memory consumption
and reduce peak memory usage by 30-40% compared to 2.0.0-beta.2
+ This release requires a clean storage directory and is not compatible
with files created by previous beta releases
- Update to v2.0.0-beta.2
+ vendor: update prometheus/tsdb (Fix panic caused by 0 division)
- Update to v2.0.0-beta.1
+ Includes a new storage layer, which reduces bottlenecks and shows
considerable performance improvements, but does not work with old v1
storage data.
- Refresh 0002-Default-settings.patch to apply cleanly again
- Add systemd unit file and default config (/etc/prometheus/prometheus.yml)
- Run as unprivileged "prometheus" user
- Default to /etc/prometheus for config and /var/lib/prometheus for metrics
+ Added 0002-Default-settings.patch
- Add BuildRequires: glibc-devel-static to fix aarch64 build
- Initial version