How is container memory allocated/shared for parallel steps?

[Deniz Cengiz]

Hi all,

I have a question regarding memory allocation for parallel steps.

Our setup looks like this:

definitions:
   steps:
      - step: &one
          name: "Step one" 
          image: <some self hosted image>
        size: 2x
        script: |
           - while true; do date && echo "Memory usage in megabytes:" && echo $((`cat /sys/fs/cgroup/memory/memory.memsw.usage_in_bytes | awk '{print $1}'`/1048576)) && echo "" && sleep 30; done >> memoryLogs.txt &
           # taken from https://support.atlassian.com/bitbucket-cloud/docs/generate-support-logs/#Container-memory-monitoring
           # some more commands
        cache: # caches being used
        artifacts:
           - memoryLogs.txt
           # some other artifacts

      - step: $two
        # similar to &one with some other scripts being executed

      - step: &three
        # ...

# ...

pipelines:
   custom:
      Run-Numbered-Steps:
         - parallel:
            - step: *one
            - step: *two
            - step: *three
            # some more parallel steps, 12 in total
         - step:
           # some other stuff that will run sequentially

From my understanding, running steps in parallel will spin up an individual container per each step, meaning in our setup we would have 12 containers running at the same time. For 2/12 of these we have defined `size: 2x` in the `definitions` section of our pipeline.yaml. So defining size 2x we would get up to 8 GB for the build container, for the ones without size: 2x half of that, so 4 GB

Now if we inspect the memory logs that are being written for each of the 12 containers (also the ones with size 2x), we always get logging along the lines of 

Mon Jun 20 12:04:51 UTC 2022
Memory usage in megabytes:
3903

Mon Jun 20 12:05:21 UTC 2022
Memory usage in megabytes:
3896

Mon Jun 20 12:05:51 UTC 2022
Memory usage in megabytes:
3917

Mon Jun 20 12:06:21 UTC 2022
Memory usage in megabytes:
3754

Mon Jun 20 12:06:51 UTC 2022
Memory usage in megabytes:
3754

Mon Jun 20 12:07:21 UTC 2022
Memory usage in megabytes:
3735

Mon Jun 20 12:07:51 UTC 2022
Memory usage in megabytes:
3740

Now, to my actual question: Are these containers actually not requiring up to 7 GB of memory to run their script commands (hence the logs with <= 4GB) or are the containers somehow only getting up to 4 GB of memory - contrary to the size 2x definition. - or is there an entirely different issue at hand?

We initially tried to increase memory for some of these steps, as the scripts running inside the steps (end-to-end tests) will give weird results and/or fail randomly, essentually becoming flaky, while on local machines they run fine. However on local machines they are granted more than 3 GB of memory, which is why we tried to also increase memory during test runs in the pipeline

Sadly the size property and memory allocation for containers in the context of parallel steps is not documented that well.

Thanks in advance - help and or directing me to the proper documentation would be highly appreciated. 

Best regards

Deniz

[Patrik S · Accepted solution]

Hello @Cengiz Deniz ,

Welcome to Atlassian Community!

Your understanding is correct and each step, be it a normal or parallel step, will spin up its own container with its own set of resources:  8GB for 2x steps and 4GB for 1x steps.

The command you are currently using to print the memory is actually printing the currently used memory and not the available memory. 

So in this case, not seeing values greater than 4GB in your log just means that your current script does not require more than 4GB of memory, although it's getting very close to that value at its peak (3903 MB in your logs), so if you limit it to 4GB (1x) you might eventually run into out of memory errors in pipelines. When the build/services container runs out of memory, the pipeline will fail and explicitly say that it's a memory-related issue.

For more details about the pipelines memory allocation you can refer to the following documentation :

• https://support.atlassian.com/bitbucket-cloud/docs/databases-and-service-containers/#Define-a-service

Thank you @Cengiz Deniz .

Kind regards,

Patrik S