Rerun Publish Steps after Network Error

[jikim-msft]

Description

665: Ability to Rerun Publish Step If There Are Any Network Error https://dev.azure.com/fluidframework/internal/_workitems/edit/665/

Cases

1. If Publish Package does not encounter an error continue with the remaining packages

2. If Publish Package encounters an error

• Network Error: retry for $maximumRetryIfNetworkError-amount of time. If network error is persisted during the entire loop, exit the task in the pipeline
• Non-Network Error: Such as the current package is already published in the registry, fail the pipeline and not continue

[curtisman]

We should fail and not continue in this case.

[jikim-msft]

Fixed it by moving code below line 140 (any error case besides the network error)

  fi
  rm "$f"
  mv "$local_f" "$f"
  exit $t1
fi

[curtisman]

This seems to exit even if the content of the package is the same?

[jikim-msft]

This is an existing behavior prior to working on the codebase. I have switched the non-network error case to the else case of the if grep -q "code ENOTFOUND" "publish_log"

[curtisman]

How do you get this error message. Did you try it in npmjs.org?

ADO feed might have a different error message. See example

[jikim-msft]

This error message existed before I started working on the codebase. The npm error message that I added is grep -q "code ENOTFOUND" which appears when there is a timeout during publishing the package to the npm registry.

[curtisman]

Please test your change to make sure it works.

[jikim-msft]

When disconnecting network in the middle of npm publish, npm ERR! code ENOTFOUND error message is returned.

Created a dummy package npm-return-error-msg which invokes the identical error message and checked that the process repeats for x-amount of times.

[curtisman]

We are still exiting even if the packages are the same?

[jikim-msft]

Added continue in line 131 to continue the loop if the packages are the same.

[curtisman]

t1 used be used for tracking how many errors we encountered, but since we are going to exist on error now, we probably don't need it anymore.

[jikim-msft]

Removed t1 from the codebase and made exit status to exit 1

As a result, made changes to the existing if-statement

Before

else
    if ! grep -q "You cannot publish over the previously published versions" "publish_log"
    then
        let t1+=1
     else
         echo Package has already been published.
          local_f="${f}_local"
          mv "$f" "$local_f"
          package_name=$(grep -oP 'npm notice package: \K.*' "publish_log")
          npm pack $package_name
          if cmp -s "$f" "$local_f"
          then
                 echo Continuing as published package matches the current one that was attempting to be released.
          else
                echo ERROR: Published package does not match the current one attempting to be released for the same version.
                 let t1+=1
         fi
      fi
    rm "$f"
    mv "$local_f" "$f"
    exit $t1
  fi

After

else
    if grep -q "You cannot publish over the previously published versions" "publish_log"
    then
         echo Package has already been published.
          local_f="${f}_local"
          mv "$f" "$local_f"
          package_name=$(grep -oP 'npm notice package: \K.*' "publish_log")
          npm pack $package_name
          if cmp -s "$f" "$local_f"
          then
                 echo Continuing as published package matches the current one that was attempting to be released.
                 continue
          else
                echo ERROR: Published package does not match the current one attempting to be released for the same version.
           fi
      fi
    rm "$f"
    mv "$local_f" "$f"
    exit $t1
  fi