acr purge in container registry no longer works. What is correct way of batch cleanup of old images

Question

acr purge in container registry no longer works. What is correct way of batch cleanup of old images

Tom D 40

We used to rely on the below task running daily. At some point it stopped running and failing with

/bin/sh: line 1: acr: command not found
2025/10/08 07:12:54 Container failed during run: acb_step_0. No retries remaining.
failed to run step ID: acb_step_0: exit status 127

is purge still supported?



```dockerfile
PURGE_CMD="acr purge \
 --filter '.*:master.*' --ago 5d --keep 10 --untagged && \
 acr purge --filter '.*:-prod.*' --ago 5d --keep 10 --untagged && \
 acr purge --filter '.*:app-prod.*' --ago 5d --keep 10 --untagged && \
 acr purge --filter '.*:release-v.*' --ago 5d --keep 15 --untagged"


az acr task update \
  --name "scheduledPurgeNonMasterTask" \
  --cmd "$PURGE_CMD" \
  --registry my-registry \
  --context /dev/null

Alex Burlachenko 18,575 Reputation points Volunteer Moderator

2025-10-30T13:13:01.5866667+00:00

Hi Tom,

the acr command you were using was part of the old acr cli tool, which has been deprecated and removed from the default container images that azure container registry tasks use.

the new, supported way to do this is to use the az acr run command with the mcr.microsoft.com/acr/acr-cli:latest container image, which contains the updated purge tool.

you need to create a new acr task that uses the new image and the new command syntax.

az acr task create --name "scheduledPurgeNonMasterTask" --registry my-registry --cmd "acr purge --filter '.:master.' --ago 5d --keep 10 --untagged" --schedule "0 0 * * *" --context /dev/null --image mcr.microsoft.com/acr/acr-cli:latest

notice the --image parameter that specifies the new cli image. you will need to create a separate task for each of your purge commands, or chain them together in a single --cmd string, just like you had before.

the reason your old task broke is that the underlying container image for the task was updated and no longer contained the old acr binary. this is a good reminder to pin to specific container image tags for critical tasks to avoid these unexpected breaks.

so, to summ. your purge logic is still correct, you just need to recreate the acr task and explicitly use the mcr.microsoft.com/acr/acr-cli:latest image to run the acr purge command.

good luck,

Alex
Himanshu Shekhar 1,860 Reputation points Microsoft External Staff Moderator

2025-11-05T10:08:49.2466667+00:00

@Tom D

Could you please provide an update on this post?

Kindly let us know if the suggested steps helps or you need further assistance on this issue.
Tom D 40 Reputation points

2025-11-05T15:50:15.8+00:00
Thanks for your suggestions, I've create another task that runs fine and succeeds, but I also re-created the old task which works but the run gets marked as a fail. Both have explicitly set the image to mcr.microsoft.com/acr/acr-cli:latest

Number of deleted tags: 8 Number of deleted manifests: 7 /bin/sh: line 1: acr: command not found 2025/11/04 07:01:42 Container failed during run: acb_step_0. No retries remaining. failed to run step ID: acb_step_0: exit status 127

So the task failed, but it still ran the script, even though it states that it fails on line 1..

Himanshu Shekhar 1,860 Microsoft External Staff Moderator

@Tom D

To fix this, always wrap multiple chained commands in a shell in your --cmd, like so: (please give a try and let us know)

az acr task create --name "scheduledPurgeNonMasterTask" \
  --registry my-registry \
  --cmd "/bin/sh -c \"acr purge --filter '.*:master.*' --ago 5d --keep 10 --untagged && acr purge --filter '.*:-prod.*' --ago 5d --keep 10 --untagged && acr purge --filter '.*:app-prod.*' --ago 5d --keep 10 --untagged && acr purge --filter '.*:release-v.*' --ago 5d --keep 15 --untagged\"" \
  --context /dev/null \
  --schedule "0 0 * * *"

Himanshu Shekhar 1,860 Reputation points Microsoft External Staff Moderator

2025-11-07T13:11:19.8533333+00:00

@Tom D - Could you please provide an update on this post?

Kindly let us know if the suggested steps helps or you need further assistance on this issue.
Tom D 40 Reputation points

2025-11-10T11:03:36.3833333+00:00

az acr task run --name "scheduledPurgeOldProductionTask" --registry my-registry

Queued a run with ID: cb13m

Waiting for an agent...

2025/11/10 11:01:57 Alias support enabled for version >= 1.1.0, please see https://aka.ms/acr/tasks/task-aliases for more information.

2025/11/10 11:01:57 Creating Docker network: acb_default_network, driver: 'bridge'

2025/11/10 11:01:57 Successfully set up Docker network: acb_default_network

2025/11/10 11:01:57 Setting up Docker configuration...

2025/11/10 11:01:58 Successfully set up Docker configuration

2025/11/10 11:01:58 Logging in to registry: my-registry.azurecr.io

2025/11/10 11:01:58 Successfully logged into my-registry.azurecr.io

2025/11/10 11:01:58 Executing step ID: acb_step_0. Timeout(sec): 600, Working directory: '', Network: 'acb_default_network'

2025/11/10 11:01:58 Launching container with name: acb_step_0

docker: invalid reference format.

See 'docker run --help'.

2025/11/10 11:01:58 Container failed during run: acb_step_0. No retries remaining.

failed to run step ID: acb_step_0: exit status 125
Tom D 40 Reputation points

2025-11-10T11:05:39.4+00:00

I also tried some variants, but this does not seem to work
Himanshu Shekhar 1,860 Reputation points Microsoft External Staff Moderator

2025-11-12T13:17:14.12+00:00
@Tom D

Try using an ACR task YAML file where you specify the image in the cmd step.

Exit code 125 is Docker's error code for invalid command syntax, returned when Docker cannot parse the arguments being passed to the docker run command.

So, when you use --image as an az acr task create parameter, the system tries to pass it directly to Docker's run command, which doesn't recognize it as a valid docker run option.

Let's try creating a file named for e.g. acr-purge-task.yaml (https://xtls-v4.hkg1.meaqua.org/en-us/azure/container-registry/container-registry-tasks-reference-yaml#acr-taskyaml-file-format)

version: v1.1.0 steps: - cmd: mcr.microsoft.com/acr/acr-cli:latest args: - /bin/sh - -c - | acr purge --filter '.*:master.*' --ago 5d --keep 10 --untagged && \ acr purge --filter '.*:-prod.*' --ago 5d --keep 10 --untagged && \ acr purge --filter '.*:app-prod.*' --ago 5d --keep 10 --untagged && \ acr purge --filter '.*:release-v.*' --ago 5d --keep 15 --untagged

Then create the task as below suggested:

az acr task create \ --name "scheduledPurgeOldProductionTask" \ --registry <registryname> \ --file <yourfilename.yaml> \ --context /dev/null \ --schedule "0 0 * * *"

After creating the task with the corrected syntax, test it:

az acr task run --name "scheduledPurgeOldProductionTask" --registry <registryname>

for e.g tried and it works adding for ref
Himanshu Shekhar 1,860 Reputation points Microsoft External Staff Moderator

2025-11-14T12:30:10.5266667+00:00

Tom D

Could you please provide an update on this post?

Kindly let us know if the suggested steps helps or you need further assistance on this issue.

1 answer

Your answer

Alex Burlachenko 18,575 Reputation points Volunteer Moderator

2025-10-30T13:13:01.5866667+00:00

Hi Tom,

the acr command you were using was part of the old acr cli tool, which has been deprecated and removed from the default container images that azure container registry tasks use.

the new, supported way to do this is to use the az acr run command with the mcr.microsoft.com/acr/acr-cli:latest container image, which contains the updated purge tool.

you need to create a new acr task that uses the new image and the new command syntax.

az acr task create --name "scheduledPurgeNonMasterTask" --registry my-registry --cmd "acr purge --filter '.:master.' --ago 5d --keep 10 --untagged" --schedule "0 0 * * *" --context /dev/null --image mcr.microsoft.com/acr/acr-cli:latest

notice the --image parameter that specifies the new cli image. you will need to create a separate task for each of your purge commands, or chain them together in a single --cmd string, just like you had before.

the reason your old task broke is that the underlying container image for the task was updated and no longer contained the old acr binary. this is a good reminder to pin to specific container image tags for critical tasks to avoid these unexpected breaks.

so, to summ. your purge logic is still correct, you just need to recreate the acr task and explicitly use the mcr.microsoft.com/acr/acr-cli:latest image to run the acr purge command.

good luck,

Alex
Himanshu Shekhar 1,860 Reputation points Microsoft External Staff Moderator

2025-11-05T10:08:49.2466667+00:00

@Tom D

Could you please provide an update on this post?

Kindly let us know if the suggested steps helps or you need further assistance on this issue.
Tom D 40 Reputation points

2025-11-05T15:50:15.8+00:00

Thanks for your suggestions, I've create another task that runs fine and succeeds, but I also re-created the old task which works but the run gets marked as a fail. Both have explicitly set the image to mcr.microsoft.com/acr/acr-cli:latest

Number of deleted tags: 8 Number of deleted manifests: 7 /bin/sh: line 1: acr: command not found 2025/11/04 07:01:42 Container failed during run: acb_step_0. No retries remaining. failed to run step ID: acb_step_0: exit status 127

So the task failed, but it still ran the script, even though it states that it fails on line 1..
Himanshu Shekhar 1,860 Reputation points Microsoft External Staff Moderator

2025-11-06T12:45:12.49+00:00

@Tom D

To fix this, always wrap multiple chained commands in a shell in your --cmd, like so: (please give a try and let us know)

az acr task create --name "scheduledPurgeNonMasterTask" \ --registry my-registry \ --cmd "/bin/sh -c \"acr purge --filter '.*:master.*' --ago 5d --keep 10 --untagged && acr purge --filter '.*:-prod.*' --ago 5d --keep 10 --untagged && acr purge --filter '.*:app-prod.*' --ago 5d --keep 10 --untagged && acr purge --filter '.*:release-v.*' --ago 5d --keep 15 --untagged\"" \ --context /dev/null \ --schedule "0 0 * * *"
Himanshu Shekhar 1,860 Reputation points Microsoft External Staff Moderator

2025-11-07T13:11:19.8533333+00:00

@Tom D - Could you please provide an update on this post?

Kindly let us know if the suggested steps helps or you need further assistance on this issue.
Tom D 40 Reputation points

2025-11-10T11:03:36.3833333+00:00

az acr task run --name "scheduledPurgeOldProductionTask" --registry my-registry

Queued a run with ID: cb13m

Waiting for an agent...

2025/11/10 11:01:57 Alias support enabled for version >= 1.1.0, please see https://aka.ms/acr/tasks/task-aliases for more information.

2025/11/10 11:01:57 Creating Docker network: acb_default_network, driver: 'bridge'

2025/11/10 11:01:57 Successfully set up Docker network: acb_default_network

2025/11/10 11:01:57 Setting up Docker configuration...

2025/11/10 11:01:58 Successfully set up Docker configuration

2025/11/10 11:01:58 Logging in to registry: my-registry.azurecr.io

2025/11/10 11:01:58 Successfully logged into my-registry.azurecr.io

2025/11/10 11:01:58 Executing step ID: acb_step_0. Timeout(sec): 600, Working directory: '', Network: 'acb_default_network'

2025/11/10 11:01:58 Launching container with name: acb_step_0

docker: invalid reference format.

See 'docker run --help'.

2025/11/10 11:01:58 Container failed during run: acb_step_0. No retries remaining.

failed to run step ID: acb_step_0: exit status 125
Tom D 40 Reputation points

2025-11-10T11:05:39.4+00:00

I also tried some variants, but this does not seem to work
Himanshu Shekhar 1,860 Reputation points Microsoft External Staff Moderator

2025-11-12T13:17:14.12+00:00

@Tom D

Try using an ACR task YAML file where you specify the image in the cmd step.

Exit code 125 is Docker's error code for invalid command syntax, returned when Docker cannot parse the arguments being passed to the docker run command.

So, when you use --image as an az acr task create parameter, the system tries to pass it directly to Docker's run command, which doesn't recognize it as a valid docker run option.

Let's try creating a file named for e.g. acr-purge-task.yaml (https://xtls-v4.hkg1.meaqua.org/en-us/azure/container-registry/container-registry-tasks-reference-yaml#acr-taskyaml-file-format)

version: v1.1.0 steps: - cmd: mcr.microsoft.com/acr/acr-cli:latest args: - /bin/sh - -c - | acr purge --filter '.*:master.*' --ago 5d --keep 10 --untagged && \ acr purge --filter '.*:-prod.*' --ago 5d --keep 10 --untagged && \ acr purge --filter '.*:app-prod.*' --ago 5d --keep 10 --untagged && \ acr purge --filter '.*:release-v.*' --ago 5d --keep 15 --untagged

Then create the task as below suggested:

az acr task create \ --name "scheduledPurgeOldProductionTask" \ --registry <registryname> \ --file <yourfilename.yaml> \ --context /dev/null \ --schedule "0 0 * * *"

After creating the task with the corrected syntax, test it:

az acr task run --name "scheduledPurgeOldProductionTask" --registry <registryname>

for e.g tried and it works adding for ref
Himanshu Shekhar 1,860 Reputation points Microsoft External Staff Moderator

2025-11-14T12:30:10.5266667+00:00

Tom D

Could you please provide an update on this post?

Kindly let us know if the suggested steps helps or you need further assistance on this issue.

Answer 1

@Tom D

Here we can suggest that the environment was updated by Azure it seems so removing the availability of the acr command unless you explicitly specify running the task with the correct container image that contains the acr-cli.

To fix this when creating or updating ACR purge tasks we must let Azure use the official acr-cli container image, which has the acr binary installed and accessible during the task runtime.

please create the task again using the official ACR CLI container image to ensure the acr command is available:

PURGE_CMD="acr purge --filter '.*:master.*' --ago 5d --keep 10 --untagged && \

acr purge --filter '.*:-prod.*' --ago 5d --keep 10 --untagged && \

acr purge --filter '.*:app-prod.*' --ago 5d --keep 10 --untagged && \

acr purge --filter '.*:release-v.*' --ago 5d --keep 15 --untagged"

then , az acr task create \

--name scheduledPurgeNonMasterTask \

--registry myRegistry \

--cmd "$PURGE_CMD" \

--schedule "0 0 * * *" \

--context /dev/null

For purging tags and manifests in ACR using az acr run with acr purge: https://xtls-v4.hkg1.meaqua.org/en-us/azure/container-registry/container-registry-auto-purge

For running an ACR task on a defined schedule : https://xtls-v4.hkg1.meaqua.org/en-us/azure/container-registry/container-registry-tasks-scheduled

Try using an ACR task YAML file where you specify the image in the cmd step.

Exit code 125 is Docker's error code for invalid command syntax, returned when Docker cannot parse the arguments being passed to the docker run command.

So, when you use --image as an az acr task create parameter, the system tries to pass it directly to Docker's run command, which doesn't recognize it as a valid docker run option.

Let's try creating a file named for e.g. acr-purge-task.yaml (https://xtls-v4.hkg1.meaqua.org/en-us/azure/container-registry/container-registry-tasks-reference-yaml#acr-taskyaml-file-format)

YAML

version: v1.1.0
steps:
  - cmd: mcr.microsoft.com/acr/acr-cli:latest
    args:
      - /bin/sh
      - -c
      - |
        acr purge --filter '.*:master.*' --ago 5d --keep 10 --untagged && \
        acr purge --filter '.*:-prod.*' --ago 5d --keep 10 --untagged && \
        acr purge --filter '.*:app-prod.*' --ago 5d --keep 10 --untagged && \
        acr purge --filter '.*:release-v.*' --ago 5d --keep 15 --untagged

Then create the task as below suggested:

Bash

az acr task create \
  --name "scheduledPurgeOldProductionTask" \
  --registry <registryname> \
  --file <yourfilename.yaml> \
  --context /dev/null \
  --schedule "0 0 * * *"

After creating the task with the corrected syntax, test it:

Bash

az acr task run --name "scheduledPurgeOldProductionTask" --registry <registryname>

for e.g tried and it works adding for ref

User's image

Share via

acr purge in container registry no longer works. What is correct way of batch cleanup of old images

1 answer

Your answer