https://flyte.org
Join Slack
hi everyone, i'm having some trouble pulling images on a sandbox deployment. everything works fine w...
# flyte-support
f
hi everyone, i'm having some trouble pulling images on a sandbox deployment. everything works fine with the base image (the one used when i started the deployment with 
flytectl sandbox start --source .
), but when i try to build another image in the sandbox (
flytectl sandbox exec -- <docker build statement
), the pod fails to pull the image even though i can see that its been successfully build via 
flytectl sandbox exec -- docker image ls
. anyone have any idea where i might be going wrong?
t
I feel like i’ve seen this recently - can you get the -o yaml dump for the pod
f
yep, just a sec
t
and grep -i imagepullpolicy
that might not matter, just curious
f
Copy code
imagePullPolicy: Always
t
can you run 
flytectl sandbox exec -- docker pull <image>
i’m hoping that fails.
👍 1
f
ah it does fail!
Copy code
Error response from daemon: repository <http://cr.flyte.org/flyteorg/tensorflow-kit|cr.flyte.org/flyteorg/tensorflow-kit> not found: name unknown: repository name not known to registry
t
that’s the image you built?
f
yep, i also tried building it without the 
<http://cr.flyte.org/flyteorg|cr.flyte.org/flyteorg>
prefix
t
oh yeah you should probably just build locally without any registry
behind the scenes it assumed dockerhub, but yeah irrelevant
um
let me do some more digging around here.
f
just as a sanity check i'll try building it with the tag 
tensorflow-kit:latest
actually, one thing that is popping up now that might be helpful is this error when i try to register the workflow:
executing:
Copy code
pyflyte register --project flytesnacks --domain development --version v1.4 --image tensorflow-kit:latest src
throws:
Copy code
AssertionError: Image Config with name tensorflow not found in the configuration
t
did you start this with flytectl sandbox? or flytectl demo
f
flytectl sandbox start --source .
t
oh it’s in the first message sorry
f
np!
i think i may not fully understand the relationship between some of the flyteadmin commands and the sandbox
t
mm
f
in this case, i have two images: 1. a base/default image 2. an image with some tensorflow stuff in it
t
can you name your image something other than latest?
the version
:tst123
f
yep sure
t
message has been deleted
f
ah interesting, let me try
while im waiting for the image to build -- how does 
pyflyte register
use the image passed to that command?
for instance, if i have a deployment running on a default/base 
image a
, but i want to register a workflow that will run on some specialized 
image b
, i should register the workflow with 
--image image-b
, right?
t
currently it’s “fast register” only. so that means the image is set in the task’s image field. separately, your code is zipped up locally and uploaded to admin (which writes it to minio). at run time, your code is downloaded from minio unzipped and then task execution runs
f
oooo
t
here
download this guy: https://blog.flyte.org/flyte-cheat-sheet the pdf
🙏 1
top right corner
multi-image workflows… that is how you can construct workflows with tasks that use two different images
though that doesn’t tell you the registration command
f
yeah i had been looking at that one
another thing im a little confused about here
t
the registration command is
Copy code
--image mydefaultimage:cafe123 --image other=nondefault:beef
f
ahhhhhh
okay that makes sense
so if my 
flyte.config
file has
Copy code
[images]
tensorflow = tensorflow-kit:v1
my register command should be:
Copy code
pyflyte register --image default:v1 tensorflow=tensorflow-kit:tst123
t
delete flyte.config
well you can delete it at least
if you don’t delete it then you need to specify it on the command line.
if you don’t specify it on the command line then 
pyflyte --config xyz …
👍 1
f
okay i just got the workflow to register with your suggestion
thanks for the help!!
🙏 1
161 Views
Open in Slack
PreviousNext
Powered by

Flyte

Flyte enables production-grade orchestration for machine learning workflows and data processing created to accelerate local workflows to production.

Powered by