The main logic with ADRs is to describe (architectural) decisions made:

* To provide best practices and solutions we (as the *build tool* team) want to promote.
* To avoid asking the same thing multiple times during code review.
* To explain *rejected solutions*, for now, and future development, in case they are proposed again.

ADRs can be written by any team.
Like code, they should be reviewed by any other relevant teams.

### Restart After Crash

But in those cases with really bad errors that crash the running **process**, you would want an external component that is in charge of **restarting** the process, at least a couple of times...

!!! tip

        // splitIntoBuckets() method expects us to split large element into N elements,
        // but we want to have a single bucket with N batches.
        // As a workaround, we repeat the bucket N times, and deduplicate the result at the end
        val resultIncludingDuplicates = splitIntoBuckets(
            LinkedList(subProjectTestClassTimes),
            SubprojectTestClassTime::totalTime,

  private var waitingCoordinatorTask: SerialTask? = null
  private var waitingCoordinatorInterrupted = false
  private var waitingCoordinatorNotified = false

  /** How many times a new task has been started. Guarded by [TaskRunner.lock]. */
  private var contextSwitchCount = 0

  /** Guarded by [TaskRunner.lock]. */
  private var activeThreads = 0

    val elapsedUntilPong3 = System.nanoTime() - startNanos
    assertThat(TimeUnit.NANOSECONDS.toMillis(elapsedUntilPong3).toDouble())
      .isCloseTo(1500.0, 250.0)

    // The client pinged the server 3 times, and it has ponged back 3 times.
    assertThat(webSocket.sentPingCount()).isEqualTo(3)
    assertThat(server.receivedPingCount()).isEqualTo(3)
    assertThat(webSocket.receivedPongCount()).isEqualTo(3)

hostnames for us to make sure that we can provide horizontal distribution, however we have come across situations where sometimes this is not feasible and there are no easier alternatives without modifying /etc/hosts on the host system as root user.  Many times in airgapped deployments this is not allowed or requires audits and approvals.

MinIO server configuration file allows users to provide topology that allows for heterogeneous hostnames, allowing MinIO to deployed in pre-existing environments...

  ): Boolean {
    task.initQueue(this)

    val now = taskRunner.backend.nanoTime()
    val executeNanoTime = now + delayNanos

    // If the task is already scheduled, take the earlier of the two times.
    val existingIndex = futureTasks.indexOf(task)
    if (existingIndex != -1) {
      if (task.nextExecuteNanoTime <= executeNanoTime) {
        taskRunner.logger.taskLog(task, this) { "already scheduled" }

    .url("https://www.nytimes.com/")
    .build();
client.newCall(newYorkTimesRequest).enqueue(new Callback() {
  ...
});
```

Running this race over home WiFi shows the Times (`0002`) completes just slightly sooner than the Post (`0001`):

```
0001 https://www.washingtonpost.com/
0001 0.000 callStart
0002 https://www.nytimes.com/
0002 0.000 callStart
0002 0.010 dnsStart

because it documents the full list of known GODEBUG settings,
which are tied to a specific release.
-->

## Introduction {#intro}

Go's emphasis on backwards compatibility is one of its key strengths.
There are, however, times when we cannot maintain complete compatibility.
If code depends on buggy (including insecure) behavior,
then fixing the bug will break that code.
New features can also have similar impacts:

- Servers running distributed MinIO instances should be less than 15 minutes apart. You can enable [NTP](http://www.ntp.org/) service as a best practice to ensure same times across servers.
- `MINIO_DOMAIN` environment variable should be defined and exported for bucket DNS style support.
- Running Distributed MinIO on **Windows** operating system is considered **experimental**. Please proceed with caution.