Menu

Important: This documentation is about an older version. It's relevant only to the release noted, many of the features and functions have been updated or replaced. Please view the current version.

Open source

V2.7

Grafana Labs is excited to announce the release of Loki 2.7. Here’s a summary of new enhancements and important fixes:

Features and enhancements

  • New Internal Server and TLS TLS can now be configured everywhere and there is an internal server to allow ring, memberlist, and readiness handlers to continue to work without TLS.
  • Better Support for Azure Blob Storage thanks to the ability to use Azure’s Service Principal Credentials.
  • Logs can now be pushed from the Loki canary so you don’t have to rely on a scraping service to use the canary.
  • Additional label_format fields __timestamp__ and __line__.
  • fifocache has been renamed The in-memory fifocache has been renamed to embedded-cache. Check upgrade guide for more details
  • New HTTP endpoint for Ingester shutdown that will also delete the ring token.
  • Faster label queries thanks to new parallization.
  • Introducing Stream Sharding an experimental new feature to help deal with very large streams.
  • Promtail
    • support for max stream limit
    • config reload endpoint / signal
    • compressed file support
    • lambda-promtail now supports Kinesis data stream events
    • matches for the journal reader
    • basic tracing support

For a full list of all, look at the CHANGELOG.

Upgrade Considerations

As always, please read the upgrade guide before upgrading Loki.

Bug fixes

2.7.5 (2023-03-28)

  • Flush buffered logger on exit: this makes sure logs are printed if Loki crashes on startup.

2.7.4 (2023-02-24)

  • Fixed different streams for cri tags ending on the same stream.
  • Fixed the userdata field (from Windows Event Log) being scraped incorrectly.
  • Fixed vector() function producing wrong timestamp.
  • Fixed behavior for overlapping chunks with multiple stores.
  • Fixed logs results caching causing query-frontend to return logs outside of query window.
  • Fixed panics when:
    • /scheduler/ring endpoint is requested with scheduler ring disabled.
    • LogQL clones a specific query.
    • Promtail deals with invalid calls to Details().

2.7.3 (2023-02-01)

  • Fixed a bug in compactor that caused divide-by-zero panics when startTime and endTime of a delete request were equal.
  • Fixed the output of the --version command that showed an incorrect version information.

2.7.2 (2023-01-25)

  • Fixed bug in validation of pattern and regexp parsers where missing or empty parameters for these parsers caused panics.
  • Fixed bugs in processing delete requests with line filters:
    • Whole chunks covered by delete requests with a line filter were incorrectly skipped.
    • Chunks partially covered by delete requests with a line filter were not properly processed. The part that was not covered by the delete request was rewritten with a line filter, while the part that was covered by the delete request was completely deleted.
  • Fixed bug in multi-tenant querying that caused HTTP 400 responses when multiple tenants where used in X-Scope-OrgID header like so tenant-a|tenant-b.
  • Upgraded Go build version and Docker container base images to 1.19.5 to mitigate GO-2022-1144 vulnerability.

2.7.1 (2022-12-09)

  • Add single compactor http client for delete and gennumber clients. This fixes a bug caused by the accidental introduction of different HTTP clients for compactor and gennumber operations that resulted in 404s when only the gennumber middlewares were enabled:

    console
    level=error ts=2022-10-18T14:13:41.598649348Z caller=delete_requests_client.go:211 msg="error getting delete requests from the store" err="unexpected status code: 404"
    ts=2022-10-18T14:13:41.598697295Z caller=spanlogger.go:80 user=application level=error msg="failed loading deletes for user" err="unexpected status code: 404"
  • Change default deletion mode to filter-only on account of the bug (see below).

Known Issues

  • There is a bug with the experimental deletion mode in 2.6.x and 2.7.x that in some cases can cause the wrong chunks to be deleted. As a result we have changed the default deletion mode to filter-only until this bug has been addressed.