search

aws / aws-lambda-base-images

Apache License 2.0
669 stars 109 forks source link

Considerations on Java base docker image size #155

Open necosta opened 6 months ago

necosta commented 6 months ago

Doing the following steps:

  1. docker pull public.ecr.aws/lambda/java:17
  2. docker run -it --entrypoint=/bin/sh public.ecr.aws/lambda/java:17
  3. cd /; du -sh

We see that this docker image has a size of 495 MB.

From AWS documentation I understand that cold starts are directly related with the size of base Java docker image so doing a further analysis on package sizes (top 10), we get:

rpm -qa --queryformat '%10{size} - %-25{name} \t %{version}\n' | sort -n | tail -10

   3639159 - bash                        4.2.46
   5395175 - dejavu-sans-fonts           2.33
   5815307 - yum                         3.4.3
   6578312 - gnupg2                      2.0.22
  12082753 - glib2                       2.56.1
  13360751 - glibc                       2.26
  14352745 - coreutils                   8.22
  25075705 - libicu                      50.2
  27436806 - python-libs                 2.7.18
 113057424 - glibc-all-langpacks         2.26

As we can see, the library that takes most space is glibc-all-langpacks (107.8 MB, 21.7% of total size)

For the majority (I guess) of lambda programs that work with default locale settings, we could replace this library with glibc-minimal-langpack (which is just a couple of KB's in size).

This would allow faster downloads and faster cold starts (when image not cached).

On https://gallery.ecr.aws/lambda/java , would you consider building an image with minimal langpack? Default image could remain with glibc-all-langpacks. Thanks.