[
https://issues.apache.org/jira/browse/BEAM-12792?focusedWorklogId=737003&page=com.atlassian.jira.plugin.system.issuetabpanels:worklog-tabpanel#worklog-737003
]
ASF GitHub Bot logged work on BEAM-12792:
-----------------------------------------
Author: ASF GitHub Bot
Created on: 05/Mar/22 02:15
Start Date: 05/Mar/22 02:15
Worklog Time Spent: 10m
Work Description: tvalentyn commented on a change in pull request #16658:
URL: https://github.com/apache/beam/pull/16658#discussion_r820015549
##########
File path: sdks/python/container/boot.go
##########
@@ -210,15 +224,37 @@ func main() {
wg.Add(len(workerIds))
for _, workerId := range workerIds {
go func(workerId string) {
- log.Printf("Executing: python %v", strings.Join(args, "
"))
- log.Fatalf("Python exited: %v",
execx.ExecuteEnv(map[string]string{"WORKER_ID": workerId}, "python", args...))
- wg.Done()
+ defer wg.Done()
+ log.Printf("Executing Python (worker %v): python %v",
workerId, strings.Join(args, " "))
+ log.Printf("Python (worker %v) exited with code: %v",
workerId, execx.ExecuteEnv(map[string]string{"WORKER_ID": workerId}, "python",
args...))
}(workerId)
}
wg.Wait()
+
+ return nil
+}
+
+// setupVenv initialize a local Python venv and set the corresponding env
variables
+func setupVenv(baseDir, workerId string) (string, error) {
+ log.Printf("Initializing temporary Python venv ...")
+
+ if err := os.MkdirAll(baseDir, 0750); err != nil {
+ return "", fmt.Errorf("Failed to create venv base directory: %s",
err)
+ }
+ dir, err := ioutil.TempDir(baseDir, fmt.Sprintf("beam-venv-%s-",
workerId))
+ if err != nil {
+ return "", fmt.Errorf("Failed Python venv directory: %s", err)
+ }
+ args := []string{"-m", "venv", "--system-site-packages", dir}
+ if err := execx.Execute("python", args...); err != nil {
+ return "", err
+ }
+ os.Setenv("VIRTUAL_ENV", dir)
Review comment:
Why are we setting these environment vars manually as opposed to
sourcing the activation script?
##########
File path: CHANGES.md
##########
@@ -130,6 +130,7 @@
* Added support for cloudpickle as a pickling library for Python SDK
([BEAM-8123](https://issues.apache.org/jira/browse/BEAM-8123)). To use
cloudpickle, set pipeline option: --pickler_lib=cloudpickle
* Added option to specify triggering frequency when streaming to BigQuery
(Python) ([BEAM-12865](https://issues.apache.org/jira/browse/BEAM-12865)).
* Added option to enable caching uploaded artifacts across job runs for Python
Dataflow jobs ([BEAM-13459](https://issues.apache.org/jira/browse/BEAM-13459)).
To enable, set pipeline option: --enable_artifact_caching, this will be
enabled by default in a future release.
+* Make Python SDK containers reusable by installing dependencies to temporary
venvs ([BEAM-12792](https://issues.apache.org/jira/browse/BEAM-12792)).
Review comment:
...reusable on Portable runners...
##########
File path: sdks/python/apache_beam/runners/worker/worker_pool_main.py
##########
@@ -94,8 +114,9 @@ def start(
# Register to kill the subprocesses on exit.
def kill_worker_processes():
+ _LOGGER.error("CLEANING UP WORKERS!!!!")
Review comment:
is this a leftover?
##########
File path: sdks/python/container/boot.go
##########
@@ -214,41 +211,105 @@ func mainError() error {
}
}
+ workerIds := append([]string{*id}, info.GetSiblingWorkerIds()...)
+
+ // Keep track of child PIDs for clean shutdown without zombies
+ childPids := struct {
+ v []int
+ canceled bool
+ mu sync.Mutex
+ } {v: make([]int, 0, len(workerIds))}
+
+ // Forward trapped signals to child process groups in order to
terminate them gracefully and avoid zombies
+ go func() {
+ log.Printf("Received signal: %v", <-signalChannel)
+ childPids.mu.Lock()
+ childPids.canceled = true
+ for _, pid := range childPids.v {
+ syscall.Kill(-pid, syscall.SIGTERM)
+ go func() {
+ // This goroutine will be canceled if the main
process exits before the 5 seconds
+ // have elapsed, i.e., as soon as all
subprocesses have returned from Wait().
+ time.Sleep(5 * time.Second)
+ log.Printf("Worker process did not respond,
killing it.")
+ syscall.Kill(-pid, syscall.SIGKILL)
+ }()
+ }
+ childPids.mu.Unlock()
+ }()
+
args := []string{
"-m",
sdkHarnessEntrypoint,
}
- workerIds := append([]string{*id}, info.GetSiblingWorkerIds()...)
var wg sync.WaitGroup
wg.Add(len(workerIds))
for _, workerId := range workerIds {
go func(workerId string) {
defer wg.Done()
+
+ childPids.mu.Lock()
+ if childPids.canceled {
+ childPids.mu.Unlock()
+ return
+ }
log.Printf("Executing Python (worker %v): python %v",
workerId, strings.Join(args, " "))
- log.Printf("Python (worker %v) exited with code: %v",
workerId, execx.ExecuteEnv(map[string]string{"WORKER_ID": workerId}, "python",
args...))
+ cmd := StartCommandEnv(map[string]string{"WORKER_ID":
workerId}, "python", args...)
+ childPids.v = append(childPids.v, cmd.Process.Pid)
+ childPids.mu.Unlock()
+
+ if err := cmd.Wait(); err != nil {
+ log.Printf("Python (worker %v) exited: %v",
workerId, err)
Review comment:
Given that this is an err branch, should this say: `Python worker
failed`, or `Python worker exited with error:...`
##########
File path: sdks/python/container/boot.go
##########
@@ -210,15 +224,37 @@ func main() {
wg.Add(len(workerIds))
for _, workerId := range workerIds {
go func(workerId string) {
- log.Printf("Executing: python %v", strings.Join(args, "
"))
- log.Fatalf("Python exited: %v",
execx.ExecuteEnv(map[string]string{"WORKER_ID": workerId}, "python", args...))
- wg.Done()
+ defer wg.Done()
+ log.Printf("Executing Python (worker %v): python %v",
workerId, strings.Join(args, " "))
+ log.Printf("Python (worker %v) exited with code: %v",
workerId, execx.ExecuteEnv(map[string]string{"WORKER_ID": workerId}, "python",
args...))
}(workerId)
}
wg.Wait()
+
+ return nil
+}
+
+// setupVenv initialize a local Python venv and set the corresponding env
variables
+func setupVenv(baseDir, workerId string) (string, error) {
+ log.Printf("Initializing temporary Python venv ...")
+
+ if err := os.MkdirAll(baseDir, 0750); err != nil {
+ return "", fmt.Errorf("Failed to create venv base directory: %s",
err)
+ }
+ dir, err := ioutil.TempDir(baseDir, fmt.Sprintf("beam-venv-%s-",
workerId))
+ if err != nil {
+ return "", fmt.Errorf("Failed Python venv directory: %s", err)
Review comment:
nit: Failed to create ... ?
##########
File path: sdks/python/container/boot.go
##########
@@ -200,25 +211,111 @@ func main() {
}
}
+ workerIds := append([]string{*id}, info.GetSiblingWorkerIds()...)
Review comment:
@robertwb @chamikaramj do we have any tests that exercise the
SiblingWorker processes functionality? IIRC, this is new. @phoerious - is it
correct that your current tests of Flink didn't test this?
##########
File path: sdks/python/container/boot.go
##########
@@ -210,15 +224,37 @@ func main() {
wg.Add(len(workerIds))
for _, workerId := range workerIds {
go func(workerId string) {
- log.Printf("Executing: python %v", strings.Join(args, "
"))
- log.Fatalf("Python exited: %v",
execx.ExecuteEnv(map[string]string{"WORKER_ID": workerId}, "python", args...))
- wg.Done()
+ defer wg.Done()
+ log.Printf("Executing Python (worker %v): python %v",
workerId, strings.Join(args, " "))
+ log.Printf("Python (worker %v) exited with code: %v",
workerId, execx.ExecuteEnv(map[string]string{"WORKER_ID": workerId}, "python",
args...))
}(workerId)
}
wg.Wait()
+
+ return nil
+}
+
+// setupVenv initialize a local Python venv and set the corresponding env
variables
Review comment:
nit: initializes and sets
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: [email protected]
For queries about this service, please contact Infrastructure at:
[email protected]
Issue Time Tracking
-------------------
Worklog Id: (was: 737003)
Time Spent: 12h 10m (was: 12h)
> Multiple jobs running on Flink session cluster reuse the persistent Python
> environment.
> ---------------------------------------------------------------------------------------
>
> Key: BEAM-12792
> URL: https://issues.apache.org/jira/browse/BEAM-12792
> Project: Beam
> Issue Type: Bug
> Components: sdk-py-harness
> Affects Versions: 2.27.0, 2.28.0, 2.29.0, 2.30.0, 2.31.0
> Environment: Kubernetes 1.20 on Ubuntu 18.04.
> Reporter: Jens Wiren
> Priority: P1
> Labels: FlinkRunner, beam
> Time Spent: 12h 10m
> Remaining Estimate: 0h
>
> I'm running TFX pipelines on a Flink cluster using Beam in k8s. However,
> extra python packages passed to the Flink runner (or rather beam worker
> side-car) are only installed once per deployment cycle. Example:
> # Flink is deployed and is up and running
> # A TFX pipeline starts, submits a job to Flink along with a python whl of
> custom code and beam ops.
> # The beam worker installs the package and the pipeline finishes succesfully.
> # A new TFX pipeline is build where a new beam fn is introduced, the pipline
> is started and the new whl is submitted as in step 2).
> # This time, the new package is not being installed in the beam worker
> causing the job to fail due to a reference which does not exist in the beam
> worker, since it didn't install the new package.
>
> I started using Flink from beam version 2.27 and it has been an issue all the
> time.
--
This message was sent by Atlassian Jira
(v8.20.1#820001)
