Github user skonto commented on a diff in the pull request:
https://github.com/apache/spark/pull/11157#discussion_r60299848
--- Diff: core/src/main/scala/org/apache/spark/util/Utils.scala ---
@@ -1978,57 +1978,134 @@ private[spark] object Utils extends Logging {
}
/**
- * Attempt to start a service on the given port, or fail after a number
of attempts.
- * Each subsequent attempt uses 1 + the port used in the previous
attempt (unless the port is 0).
- *
- * @param startPort The initial port to start the service on.
- * @param startService Function to start service on a given port.
- * This is expected to throw java.net.BindException
on port collision.
- * @param conf A SparkConf used to get the maximum number of retries
when binding to a port.
- * @param serviceName Name of the service.
- * @return (service: T, port: Int)
- */
- def startServiceOnPort[T](
- startPort: Int,
+ * Attempt to start a service on the given port, or fail after a number
of attempts.
+ * Each subsequent attempt uses 1 + the port used in the previous
attempt (unless the port is 0).
+ * It takes into consideration port restrictions through the env var
AVAILABLE_PORTS
+ *
+ * @param startPort The initial port to start the service on.
+ * @param startService Function to start service on a given port.
+ * This is expected to throw java.net.BindException
on port collision.
+ * @param conf A SparkConf used to get the maximum number of retries
when binding to a port.
+ * @param serviceName Name of the service.
+ * @return (service: T, port: Int)
+ */
+ def startServiceOnPort[T](startPort: Int,
startService: Int => (T, Int),
conf: SparkConf,
serviceName: String = ""): (T, Int) = {
+ val serviceString = if (serviceName.isEmpty) "" else s" '$serviceName'"
+
+ // define some helpers, they all share common data, maybe a service
abstract class
+ // for all services could be a good fit here.
+
+ def portRangeToList(ranges: String): List[(Long, Long)] = {
+ if (ranges == "") {
+ return List()
+ }
+ ranges.split(" ").map { r => val ret = r.substring(1, r.length -
1).split(",")
+ (ret(0).toLong, ret(1).toLong)
+ }.toList
+ }
+
+ def startOnce(tryPort: Int): (Option[T], Int) = {
+ val serviceString = if (serviceName.isEmpty) "" else s"
'$serviceName'"
+ try {
+ val (service, port) = startService(tryPort)
+ logInfo(s"Successfully started service$serviceString on port
$port.")
+ (Some(service), port)
+ } catch {
+ case e: Exception if isBindCollision(e) =>
logWarning(s"Service$serviceString " +
+ s"could not bind on port $tryPort. ")
+ (None, -1)
+ }
+ }
+
+ def retryPort(next: Int => Int, maxRetries: Int): (T, Int) = {
+
+ for (offset <- 0 until maxRetries) {
+ val tryPort = next(offset)
+ try {
+ val (service, port) = startService(tryPort)
+ logInfo(s"Successfully started service$serviceString on port
$port.")
+ return (service, port)
+ } catch {
+ case e: Exception if isBindCollision(e) =>
+ if (offset >= maxRetries) {
+ val exceptionMessage =
+ s"${e.getMessage}: Service$serviceString failed after
$maxRetries retries!"
+ val exception = new BindException(exceptionMessage)
+ // restore original stack trace
+ exception.setStackTrace(e.getStackTrace)
+ throw exception
+ }
+ logWarning(s"Service$serviceString could not bind on port
$tryPort.")
+ }
+ }
+ // Should never happen
+ throw new SparkException(s"Failed to start service$serviceString on
port $startPort")
+ }
+
+ def startFromAvailable(rand: Boolean = false): (T, Int) = {
+ val ports = portRangeToList(sys.env.get("AVAILABLE_RAND_PORTS").get)
--- End diff --
It is called when AVAILABLE_RAND_PORTS is set. So if spark.executor.port or
spark.blockManager.port are set to a specific value then this will be used so
no collision...
Again... if they are not set.. startFromAvailable(true) ->
retryPort(tryPort, maxRetries) will be called . The former shuffles the ports
and pass them down to the latter. Also retryPort uses the old logic where each
port will be tried until there is one free.
Another approach is to preassign
spark.executor.port,spark.blockManager.port with some offered values explicitly
and carry those values with AVAILABLE_RAND_PORTS.
This env variable is a clean way to do it i think without changing a lot
the code base.
---
If your project is set up for it, you can reply to this email and have your
reply appear on GitHub as well. If your project does not have this feature
enabled and wishes so, or if the feature is enabled but not working, please
contact infrastructure at [email protected] or file a JIRA ticket
with INFRA.
---
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
