为什么我在使用 Dataflow 管道时遇到“错误同步 pod”？

Question

I experiment a weird error with my Dataflow pipeline when I want to use specific library from PyPI.

I need jsonschema in a ParDo, so, in my requirements.txt file, I added jsonschema==3.2.0 . I launch my pipeline with the command line below:

python -m gcs_to_all \
    --runner DataflowRunner \
    --project <my-project-id> \
    --region europe-west1 \
    --temp_location gs://<my-bucket-name>/temp/ \
    --input_topic "projects/<my-project-id>/topics/<my-topic>" \
    --network=<my-network> \
    --subnetwork=<my-subnet> \
    --requirements_file=requirements.txt \
    --experiments=allow_non_updatable_job \
    --streaming  

In the terminal, all seems to be good:

INFO:root:2020-01-03T09:18:35.569Z: JOB_MESSAGE_BASIC: Worker configuration: n1-standard-4 in europe-west1-b.
INFO:root:2020-01-03T09:18:35.806Z: JOB_MESSAGE_WARNING: The network default doesn't have rules that open TCP ports 12345-12346 for internal connection with other VMs. Only rules with a target tag 'dataflow' or empty target tags set apply. If you don't specify such a rule, any pipeline with more than one worker that shuffles data will hang. Causes: Firewall rules associated with your network don't open TCP ports 12345-12346 for Dataflow instances. If a firewall rule opens connection in these ports, ensure target tags aren't specified, or that the rule includes the tag 'dataflow'.
INFO:root:2020-01-03T09:18:48.549Z: JOB_MESSAGE_DETAILED: Workers have started successfully.

Where's no error in the log tab on Dataflow webpage, but in stackdriver:

message: "Error syncing pod 6515c378c6bed37a2c0eec1fcfea300c ("<dataflow-id>--01030117-c9pc-harness-5lkv_default(6515c378c6bed37a2c0eec1fcfea300c)"), skipping: [failed to "StartContainer" for "sdk0" with CrashLoopBackOff: "Back-off 10s restarting failed container=sdk0 pod=<dataflow-id>--01030117-c9pc-harness-5lkv_default(6515c378c6bed37a2c0eec1fcfea300c)""
message: ", failed to "StartContainer" for "sdk1" with CrashLoopBackOff: "Back-off 5m0s restarting failed container=sdk1 pod=<dataflow-id>--01030117-c9pc-harness-5lkv_default(6515c378c6bed37a2c0eec1fcfea300c)"" 
message: ", failed to "StartContainer" for "sdk2" with CrashLoopBackOff: "Back-off 5m0s restarting failed container=sdk2 pod=<dataflow-id>--01030117-c9pc-harness-5lkv_default(6515c378c6bed37a2c0eec1fcfea300c)"" 
message: ", failed to "StartContainer" for "sdk3" with CrashLoopBackOff: "Back-off 5m0s restarting failed container=sdk3 pod=<dataflow-id>--01030117-c9pc-harness-5lkv_default(6515c378c6bed37a2c0eec1fcfea300c)"" 

I find this error too (in info mode):

Collecting jsonschema (from -r /var/opt/google/staged/requirements.txt (line 1))
  Installing build dependencies: started
Looking in links: /var/opt/google/staged
  Installing build dependencies: started
Collecting jsonschema (from -r /var/opt/google/staged/requirements.txt (line 1))
  Installing build dependencies: started
Looking in links: /var/opt/google/staged
Collecting jsonschema (from -r /var/opt/google/staged/requirements.txt (line 1))
  Installing build dependencies: started
  Installing build dependencies: finished with status 'error'
  ERROR: Command errored out with exit status 1:
   command: /usr/local/bin/python3 /usr/local/lib/python3.7/site-packages/pip install --ignore-installed --no-user --prefix /tmp/pip-build-env-mdurhav9/overlay --no-warn-script-location --no-binary :none: --only-binary :none: --no-index --find-links /var/opt/google/staged -- 'setuptools>=40.6.0' wheel
       cwd: None
  Complete output (5 lines):
  Looking in links: /var/opt/google/staged
  Collecting setuptools>=40.6.0
  Collecting wheel
    ERROR: Could not find a version that satisfies the requirement wheel (from versions: none)
  ERROR: No matching distribution found for wheel

But I don't know why it can get this dependency...

Do you have any idea how I can debug this? or why I encounter this error?

Thanks

Answer 1

When Dataflow workers start, they execute several steps:

1. Install packages from requirements.txt
2. Install packages specified as extra_packages
3. Install workflow tarball and execute actions provided in setup.py .

Error syncing pod with CrashLoopBackOff message can be related to dependency conflict. You need to verify that there are no conflicts with the libraries and versions used for the job. Please refer to the documentation for staging required dependencies of the pipeline.

Also, take a look for preinstalled dependencies and this StackOverflow thread .

What you can try is change the version of jsonschema and try run it again. If it wouldn't help, please provide requirements.txt file.

I hope it will help you.

Answer 2

There is a playbook for this error: https://cloud.google.com/dataflow/docs/guides/common-errors#error-syncing-pod