1. Proposed title of this feature request

When the network is unstable, crio is unable to pull images from docker.io while docker cli can. It makes OCP4 less tolerant to network outage than OCP3

2. What is the nature and description of the request?

Docker cli has a retry mechanism per blob which is very efficient which is not the case with crio stack. 

There is an RFE upstream https://github.com/containers/image/issues/1145

Example on customer's infrastructure

• Pull with buildah --> KO

1. https_proxy=<obfuscated> buildah pull docker://docker.io/image:tag
   Getting image source signatures
   Copying blob 04f220ee9266 done
   Copying blob 034655750c88 done
   Copying blob f0b757a2a0f0 done
   Copying blob 89c8a77f7842 done
   Copying blob 4bbcce26bc5e done
   Copying blob 7b1a6ab2e44d done
   Copying blob d1de5652303b done
   Copying blob ef669123e59e done
   Copying blob b14b1ba1d651 done
   Copying blob e5cec468d3a6 done
   while pulling "docker://docker.io/image:tag" as "docker.io/image:tag": Error writing blob: error storing blob to file "/var/tmp/storage487953815/6": read tcp x.x.x.x:port->x.x.x.x:3128: read: connection reset by peer

• Pull with docker --> OK

1. https_proxy=<obfuscated> docker pull docker.io/image:tag
   Trying to pull repository docker.io/library/image ...
   10.6.5: Pulling from docker.io/library/image
   7b1a6ab2e44d: Pull complete
   034655750c88: Pull complete
   f0b757a2a0f0: Pull complete
   4bbcce26bc5e: Pull complete
   04f220ee9266: Pull complete
   89c8a77f7842: Pull complete
   d1de5652303b: Pull complete
   ef669123e59e: Pull complete
   e5cec468d3a6: Pull complete
   b14b1ba1d651: Pull complete
   Status: Downloaded newer image for docker.io/image:tag

Docker logs show that it has to retry a blob when pulling the image

1. journalctl -u docker | grep resume
   dockerd-current[pid]: time="" level=debug msg="attempting to resume download of \"sha256:7b1a6ab2e44d...\" from 653653 bytes"

3. Why does the customer need this? (List the business requirements here)

The customer is planning the migration from OCP 3.11 to 4.x but this issue make them worried about the cluster/applications availability when the network is getting unstable or the company proxy drops a connection.

4. List any affected packages or components.

Crio stack on CoreOS 4.9/RHEL 8.4