This finally adds support for multiple architectures! 🎉 

Images are now built using BuildKit (actually img), Dockerfiles were parameterized to allow this.
Containers are still built on alpine, using QEMU and binfmt as a kernel compatibility layer to allow RUN steps in these.

CircleCI has been fully revamped, to speed up builds and de-duplicate the configuration.
Makefile has been extended with additional logic to shadow the multi-arch to non-CI users.

Happy hacking on Raspberry Pi and friends :)

* Add release namespace for loki

* Add Release namespace for promtail chart

* Add Release namespace for loki-stack chart

* Add doumentation

* Increase chart version

Logcli used to inconsistently handle timezones in timestamps.

This addresses this by adding a new `--timezone` / `-z` flag, to explicitly set the timezone the timestamps shall be formatted in.

Furthermore it refactors the output framework inside of logcli to test it easier.

Fixes #798 

* Use single quote to avoid escaping regex

Adds an example of how to setup the `GRAFANA_ADDR` environment variable when not using Grafana Cloud but rather a local or otherwise hosted instance of Loki

* fix tail library logs to use our own log format

* PR Feedbacks

docker-driver-push does a build of all the go files and as such needs to make called with BUILD_IN_CONTAINER=false from circleci

refactor(makefile): simplify makefile by removing the autogenerated targets and make building binaries and dockerfiles separate operations.

Registering the Ingester service to gRPC enables the chunk transfer
mechanism to work.

PR #791 accidentally removed the ability to use the promtail UI
by removing methods from the Target interface that were only
used within the HTML templates.

Along with restoring the methods in the Target interface, this
commit also introduces details for the JournalTarget, which
currently only provides the position in the journal being
tracked.

ingester: support chunk transfers on ingester shutdown.

add a Name() method to the stage interface so that debug logging can show you the name of the pipeline stage which just processed the log
remove some unnecessary logging around fsnotify events we don't care about and saving positions
Make the Processing Log Lines doc a first class citizen, I reference this a lot and currently it's hidden behind 3 clicks

This commit introduces chunk transfers, borrowing the mechanism from
Cortex's implementation: when an ingester is shut down with claim
on rollout enabled, the ingester will find a pending ingester and
transfer all of its chunks to it.

Document centos dep

Signed-off-by: Xiang Dai <764524258@qq.com>

- document centos dep
- add `-y` option

Signed-off-by: Xiang Dai <764524258@qq.com>

* Use prometheus pool for line buffer.

* refactor(querier/ingester): TailRequest Lookback window

Moves the specifications of the Lookback Window out of the
logproto.QueryRequest into it's own type logproto.Lookback.
This is required, because the Lookback Window will be used in the TailRequest
as well.

* feat(querier): parse Lookback from HTTP Request

* feat(logcli): send Lookback Window spec with tail request

* feat(querier): include historic entries in tail mode

Extends tailing by sending a configurable amount of historic entries with
before the live entries. This enables a behaviour that is closer to kubectl logs
-f and docker logs -f.

It is implemented by running a regular Query before subscribing to the ingesters.

* fix: adapt tests to Lookback change

* feat(querier): check all errors to make the linter happy

* fix(logproto): flatten Lookback window spec

Flattens the Lookback window spec into the individual queries

* fix(ingester): adapt test to Lookback flatten

Fixes #774

* rounds nanoseconds boundaries to milliseconds

* convert also store query

moving loki-canary project under loki repo