3 Commits

Author SHA1 Message Date
Edwin Kempin
d92f6f754a Add support for logging tags
By opening a TraceContext it is now possible to set a logging tag that
will be automatically included into all subsequent log statements that
are done by the same thread. When the TraceContext is closed the tag is
unset.

A tag has a name and a value. Having multiple tags with the same name
but different values is possible.

Adding a tag with the same name and same value a second time is possible
but has no effect.

With this change open TraceContexts are not propagated to newly started
threads that do work in the background. Support for this will be
implemented by the follow-up change.

Example:

  // Initially there are no tags
  logger.atSevere().log("log without tag");

  // a tag can be set by opening a TraceContext
  try (TraceContext ctx = new TraceContext("tag1", "value1")) {
    logger.atSevere().log("with tag1=value1");

    // by opening another TraceContext a another tag can be added
    try (TraceContext ctx2 = new TraceContext("tag2", "value2")) {
      logger.atSevere().log("log with tag1=value1 and tag2=value2");

      // it's possible to have same tag name with multiple values
      try (TraceContext ctx3 = new TraceContext("tag2", "value2a")) {
        logger.atSevere().log("log with tag1=value1, tag2=value2 and tag2=value2a");

        // adding a tag with the same name and value of an existing tag has no effect
        try (TraceContext ctx4 = new TraceContext("tag2", "value2a")) {
          logger.atSevere().log("log with tag1=value1, tag2=value2 and tag2=value2a");
        }

        // closing ctx4 didn't remove tag2=value2a since it was set before opening ctx4
        logger.atSevere().log("log with tag1=value1, tag2=value2 and tag2=value2a");
      }

      // closing ctx3 removed tag2=value2a
      logger.atSevere().log("log with tag1=value1 and tag2=value2");
    }

    // closing ctx2 removed tag2=value2
    logger.atSevere().log("with tag1=value1");
  }

  // closing ctx1 removed tag1=value1
  logger.atSevere().log("log without tag");

To support logging tags we must register a LoggingContext in Flogger.
This is done by setting the 'flogger.logging_context' system property.

Change-Id: Ib01eb90b9ad832be0b2d794e14ce23bdef2f0fb6
Signed-off-by: Edwin Kempin <ekempin@google.com>
2018-08-14 08:07:59 +02:00
Edwin Kempin
2b81d71374 Add dependency to Flogger
Flogger is an open source project (Apache 2 License) from Google that
provides a fluent JAVA API for logging:
* https://github.com/google/flogger
* https://google.github.io/flogger/

We want to use Flogger instead of slf4j. This idea was discussed at the
Gerrit hackathon in Lund [1] and on the mailing list [2].

Benefits:
* Better Readability
* Better Performance
* Will allow us to implement request tracing

Flogger API examples:
* logger.atSevere()
    .withCause(err)
    .log("Error in %s %s", req.getMethod(), uri);
* logger.atInfo()
    .atMostEvery(1, TimeUnit.HOURS)
    .log(...)
* logger.atInfo()
    .withStackTrace(StackSize.FULL)
    .log(...)

Within Google Flogger is long-established and is used by almost 3/4 of
all Google's Java projects, hence this library is very mature.

The Flogger API can be used with different logging backends. We are
using the log4j backend, as we are currently using log4j as backend with
slf4j. This means there are no changes to how logs are configured
(existing log4j.properties and log4j.xml files continue to work). It
also means that this doesn't effect any integrations that work with
log4j (e.g. integration with logstash continues to work).

To use the log4j backend for Flogger the system property
'flogger.backend_factory' must be set to
'com.google.common.flogger.backend.log4j.Log4jBackendFactory#getInstance'.
E.g. this system property may be set in gerrit.config:
[container]
  javaOptions = "-Dflogger.backend_factory=com.google.common.flogger.backend.log4j.Log4jBackendFactory#getInstance"

If this property is not set Gerrit sets it automatically on startup so
that we always have a proper logging configuration. We do this in
Main.java before loading any class. This is important since Flogger
reads this system property only once when the FluentLogger class is
loaded and we must be sure that the system property was set before this.

Using Flogger and slf4j in parallel is not a problem. Logs from both
APIs end up in the same log file(s).

This change doesn't migrate any classes to Flogger yet. This will be
done by follow-up changes.

JGit is still using slf4j and needs an explicit runtime dependency on
slf4j.

[1] https://docs.google.com/presentation/d/1Kwn77c1PScMLDEWiJDklEm43NsN2-mWm2PqHwYTkJfI/edit?usp=sharing
[2] https://groups.google.com/d/msg/repo-discuss/meEWJO7c4pA/fxDx2waRCQAJ

Change-Id: Ifc31fb95c10397857e8de5657c729f49c1820ec4
Signed-off-by: Edwin Kempin <ekempin@google.com>
2018-06-05 13:06:05 +02:00
David Ostrovsky
7cd8475275 Dissolve gerrit-main top-level directory
Change-Id: I9c217f640bbc94fc594280e3bd06683ea4ec978d
2017-10-31 11:02:36 -04:00