Change cache key calculation to be more reproducible. #525
Conversation
|
Thanks! I've tested this image with a real-world example (which I previously reduced to the example from #521), and for some reason caching still doesn't work. I've shared the full example at https://github.com/broadinstitute/kaniko-stage-caching-bug/blob/master/Dockerfile.full Could you take a look? I'm also still trying to understand how to enable more verbose logs. |
Definitely! I added some better logging in this PR so once it goes in it should be easier to debug cache misses like this. |
| RUN date > /date | ||
| COPY context/foo /foo | ||
| RUN echo hey | ||
| FROM alpine as base_stage |
There was a problem hiding this comment.
Would this Dockerfile fail integration tests if caching didn't work? As in, if kaniko built it a second time without the cache during the tests, I'm not sure if container-diff would catch the difference (that's why I had RUN date > /date in the old version, since container-diff would fail if the date layer wasn't pulled from the cache).
There was a problem hiding this comment.
Great point! I changed this to reproduce the caching issue I was hitting, but dropped the date thing. I'll double check.
Before we were using the full image digest, but that contains a timestamp. Now we only use the layers themselves and the image config (env vars, etc.). Also fix a bug in unpacking the layers themselves. mtimes can change during unpacking, so set them all once at the end.
…ContainerTools#525)" This reverts commit 1ffae47.
Before we were using the full image digest, but that contains a timestamp. Now
we only use the layers themselves and the image config (env vars, etc.).
Also fix a bug in unpacking the layers themselves. mtimes can change during unpacking,
so set them all once at the end.
I think this will fix #521