Retry dataproc transient errors

[OSalama]

resolves #1271
docs dbt-labs/docs.getdbt.com/# N/A

Problem

Python models that are configured to use dataproc serverless jobs will fail when the polling API call returns a transient 5xx error. The actual dataproc batch job will still run to completion successfully, and data becomes available in the table, but all downstream models will get skipped because dbt thinks it failed.

Solution

This PR solves the problem by adding exponential retry to the API polling calls, which will retry all transient API errors.

It also increases the time between polling calls while the dataproc job is still in a "pending" state. This is because Dataproc serverless batch jobs take at least 60 seconds to start up, which leads to a lot of additional polling, and increases the chance of getting a transient error.
This increase in time does mean that any models that normally complete within 10 seconds of the Dataproc job transitioning from "pending" to "running" will see increased runtime. In practice I would be very surprised if this were to cause any issues, as this is only impacting jobs that are submitted via dataproc serverless, which are expected to run longer than interactive/dedicated cluster methods.

I have run this in our production environment for 2 weeks with no further issues.
I generated a patch file by checking out 9efeb859d1ac21cab1bb6441acc06ec1328e4888 (release 1.8.1), modifying batch.py with my changes, then git diff batch.py > batch.py.patch
I then use the below Dockerfile to apply the patch on top of Python 3.9.8 & dbt-bigquery 1.8.1.

FROM python:3.9.8-slim-buster
RUN apt-get update -q && apt-get install -q -y patch
RUN pip install dbt-bigquery==1.8.1
COPY docker/batch.py.patch /tmp/batch.py.patch
RUN patch /usr/local/lib/python3.9/site-packages/dbt/adapters/bigquery/dataproc/batch.py < /tmp/batch.py.patch

Checklist

• I have read the contributing guide and understand what's expected of me
• I have run this code in development and it appears to resolve the stated issue
• This PR includes tests, or tests are not required/relevant for this PR
• This PR has no interface changes (e.g. macros, cli, logs, json artifacts, config files, adapter interface, etc) or this PR has already received feedback and approval from Product or DX

[cla-bot]

Thank you for your pull request and welcome to our community. We could not parse the GitHub identity of the following contributors: Omar Salama.
This is most likely caused by a git client misconfiguration; please make sure to:

1. check if your git client is configured with an email to sign commits git config --list | grep email
2. If not, set it up using git config --global user.email [email protected]
3. Make sure that the git commit email is configured in your GitHub account settings, see https://github.com/settings/emails

[OSalama]

I am finding it hard to reproduce the issue in a test environment where I'm just running a handful of Python models, but the production environment hits the problem daily (10s/100s of concurrent Python models). I'll be applying this patch to our prod environment for our next daily run, and if all goes well, I'll check off the relevant testing checklist items.

[OSalama]

I've been running this in our production environment for a couple weeks and have not seen any more cases of Python models incorrectly marked as failed.

I have run this in our production environment for 2 weeks with no further issues.
I generated a patch file by checking out 9efeb859d1ac21cab1bb6441acc06ec1328e4888 (release 1.8.1), modifying batch.py with my changes, then git diff batch.py > batch.py.patch
I then use the below Dockerfile to apply the patch on top of Python 3.9.8 & dbt-bigquery 1.8.1.

FROM python:3.9.8-slim-buster
RUN apt-get update -q && apt-get install -q -y patch
RUN pip install dbt-bigquery==1.8.1
COPY docker/batch.py.patch /tmp/batch.py.patch
RUN patch /usr/local/lib/python3.9/site-packages/dbt/adapters/bigquery/dataproc/batch.py < /tmp/batch.py.patch