Skip to content

Document approaches for site builds on top of EESSI - #778

Open
casparvl wants to merge 32 commits into
EESSI:mainfrom
casparvl:site_builds
Open

casparvl wants to merge 32 commits into
EESSI:mainfrom
casparvl:site_builds

Conversation

@casparvl

@casparvl casparvl commented Jun 4, 2026

Copy link
Copy Markdown
Collaborator

No description provided.

@casparvl casparvl changed the title Add initial files for each site-building approach Document approaches for site builds on top of EESSI Jun 4, 2026

@ocaisa ocaisa left a comment

Copy link
Copy Markdown
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Found some typos

Comment thread docs/site_build/overview.md Outdated
Comment thread docs/site_build/overview.md Outdated
Comment thread docs/site_build/overview.md Outdated
Comment thread docs/site_build/overview.md Outdated
Comment thread docs/site_build/overview.md Outdated
Comment thread docs/site_build/site_cvmfs.md Outdated
Comment thread docs/site_build/site_cvmfs.md Outdated
Comment thread docs/site_build/site_cvmfs.md Outdated
Comment thread docs/site_build/site_cvmfs.md
@casparvl
casparvl marked this pull request as ready for review June 30, 2026 14:36
Caspar van Leeuwen and others added 2 commits July 1, 2026 17:36
…cript. Also, added a full, copyable script combining all the previous components.

@Neves-P Neves-P left a comment

Copy link
Copy Markdown
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Looks really nice! This is my first pass of comments, as of know I only focused on the text. This week I will follow the documentation in practice and review again.


Another thing to consider is to create a secondary user with `aws iam create-user --user-name <name>` and attach a very limited policy to it (e.g. only read/write/list on buckets, nothing else). Then, create credentials for this user with `aws iam create-access-key --user-name <name>` and provide those credentials to the EESSI build bot and Stratum 0 machines. That way, if that token is compromized, the impact is minimized (e.g. the token can at least not be used to create new IAM idententies, etc).

## Setting up the EESSI build bot

Copy link
Copy Markdown
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I'd consider separating this into a new page. The build bot is somewhat self contained and the page is quite long already.

Comment thread docs/site_build/shared_fs.md Outdated
Comment thread docs/site_build/shared_fs.md Outdated
Comment thread docs/site_build/shared_fs.md Outdated
Comment thread docs/site_build/site_cvmfs.md
Comment thread docs/site_build/site_cvmfs.md Outdated
Comment thread docs/site_build/site_cvmfs.md Outdated

Another thing to consider is to create a secondary user with `aws iam create-user --user-name <name>` and attach a very limited policy to it (e.g. only read/write/list on buckets, nothing else). Then, create credentials for this user with `aws iam create-access-key --user-name <name>` and provide those credentials to the EESSI build bot and Stratum 0 machines. That way, if that token is compromised, the impact is minimized (e.g. the token can at least not be used to create new IAM idententies, etc).

## Setting up the EESSI build bot

Copy link
Copy Markdown
Collaborator Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I realize these docs lack one crucial thing: setting up a site_config_script that sets EESSI_SITE_INSTALL_FORCE and EESSI_SITE_SOFTWARE_PATH_PREFIX

@casparvl casparvl Jul 22, 2026

Copy link
Copy Markdown
Collaborator Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I.e. at SURF, our site_config_script looks like this:

# To build on top of EESSI, we need to software.eessi.io repository to be mounted next to our own repository
# The bot/build.sh script does this when the EESSI_SITE_INSTALL_FORCE environment variable is set
# Other build scripts will also respect this variable where needed in order to make sure that 'building on top'
# of EESSI is possible
export EESSI_SITE_INSTALL_FORCE=1
echo "Value of EESSI_SITE_INSTALL_FORCE: $EESSI_SITE_INSTALL_FORCE"

# We also need to set a prefix that our installations should end up in
# The build scripts should take this prefix, and construct the final EESSI_SITE_SOFTWARE_PATH out of it
# that the EESSI-extend module expects
export EESSI_SITE_SOFTWARE_PATH_PREFIX=/cvmfs/software.surf.nl/versions/2025.06/
echo "Value of EESSI_SITE_SOFTWARE_PATH_PREFIX: $EESSI_SITE_SOFTWARE_PATH_PREFIX"

One thing to figure out might be how to avoid hard-coding the version in here :)

Copy link
Copy Markdown
Collaborator Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Scratch that, this should not be needed anymore... I think we do this differently now... We automatically set these here https://github.com/EESSI/software-layer-scripts/blob/ca929cd7ef32a9fcd79fafd4e0d5c362a1fff452/bot/build.sh#L156 . That happens if the repo_name in the repos config file is unequal to software.eessi.io or dev.eess.io essentially.

@Neves-P Neves-P Aug 4, 2026

Copy link
Copy Markdown
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I think this does warrant a small explanation of why EESSI_SITE_SOFTWARE_PREFIX, EESSI_SITE_INSTALL and friends are not needed with this workflow. Maybe we don't even need an explanation, but just a mention.

I'm saying this because going through the whole docs - especially the shared file system installs - you get the impression that at least doing EESSI_SITE_INSTALL is necessary for site installations (which it is, but that happens behind the scenes).

@Neves-P Neves-P left a comment

Copy link
Copy Markdown
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Just two comments, suggestions. I would have liked to have been more thorough, but overall I did two passes to the text and managed to get to the point of installing the bot (had to wait a bit for the internal S3 and maintenance).

In my mind, after addressing these comments we can either wait a little longer and I spin up a new test CVMFS infrastructure to also test that part, or merge before that and address any thing else we spot in a later PR.

Overall I think the instructions are very nice! I'm not sure I am able to put myself in the shoes of a complete novice to EESSI, but they seem very appropriate for that target audience.


Another thing to consider is to create a secondary user with `aws iam create-user --user-name <name>` and attach a very limited policy to it (e.g. only read/write/list on buckets, nothing else). Then, create credentials for this user with `aws iam create-access-key --user-name <name>` and provide those credentials to the EESSI build bot and Stratum 0 machines. That way, if that token is compromised, the impact is minimized (e.g. the token can at least not be used to create new IAM idententies, etc).

## Setting up the EESSI build bot

@Neves-P Neves-P Aug 4, 2026

Copy link
Copy Markdown
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I think this does warrant a small explanation of why EESSI_SITE_SOFTWARE_PREFIX, EESSI_SITE_INSTALL and friends are not needed with this workflow. Maybe we don't even need an explanation, but just a mention.

I'm saying this because going through the whole docs - especially the shared file system installs - you get the impression that at least doing EESSI_SITE_INSTALL is necessary for site installations (which it is, but that happens behind the scenes).

``` { .bash .copy }
curl "https://awscli.amazonaws.com/awscli-exe-linux-x86_64.zip" -o "awscliv2.zip"
unzip awscliv2.zip
sudo ./aws/install

Copy link
Copy Markdown
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

This can be installed in userland, and ideally I'd do so. That's how I normally install it in the dev.eessi.io bots and it works well.

Comment thread docs/site_build/site_cvmfs.md Outdated

@Neves-P Neves-P left a comment

Copy link
Copy Markdown
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Another small detail that I spotted now and am adding here so I don't forget.

Comment thread docs/site_build/site_cvmfs.md Outdated
@Neves-P

Neves-P commented Sep 1, 2026

Copy link
Copy Markdown
Member

The last commit message was supposed to read: "Account for custom prefix installs for .cvmfsdirtab"

After that is reviewed, we need to account for the changes in the tarball ingestion step too. That is blocked until EESSI/filesystem-layer#278 is merged.

Comment thread docs/site_build/site_cvmfs.md Outdated
bedroge and others added 2 commits September 2, 2026 09:53
Co-authored-by: Pedro Santos Neves <10762799+Neves-P@users.noreply.github.com>
Co-authored-by: Bob Dröge <b.e.droge@rug.nl>
Comment thread docs/site_build/site_cvmfs.md Outdated
Co-authored-by: Bob Dröge <b.e.droge@rug.nl>

@Neves-P Neves-P left a comment

Copy link
Copy Markdown
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Suggestions with updates accounting for VERSIONS_SUBPATH workflow in ingest-tarball.sh.

DOWNLOAD_DIR=/prefix/for/tarball/staging # Some directory to temporarily store tarballs on the Stratum 0
ALLOWED_SIGNERS=/path/to/allowed/signers/file # Optional, needed in step 4
REPO_NAME="<repo_name>"
# a name for a dir in the bucket in which to archive tarballs, so that a subsequent run doesn't re-ingest them

Copy link
Copy Markdown
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Suggested change
# a name for a dir in the bucket in which to archive tarballs, so that a subsequent run doesn't re-ingest them
# Repository-relative path to the versions directory.
# This must match the versions_subpath used for .cvmfsdirtab.
VERSIONS_SUBPATH="versions"
# For a site prefix such as /cvmfs/name.sitename.tld/eessi, use:
# VERSIONS_SUBPATH="eessi/versions"
# a name for a dir in the bucket in which to archive tarballs, so that a subsequent run doesn't re-ingest them

# 3. Update the lmod caches for your site installation prefix
INGEST_SCRIPT="$PWD/filesystem-layer/scripts/ingest-tarball.sh"
```

Copy link
Copy Markdown
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Suggested change
`VERSIONS_SUBPATH` must match the `versions_subpath` used when [configuring `.cvmfsdirtab`](#setting-up-your-stratum-0). It is relative to the root of the target CVMFS repository and must include the `versions` directory.
For example, with `EESSI_SITE_SOFTWARE_PREFIX=/cvmfs/name.sitename.tld/eessi`, use `VERSIONS_SUBPATH=eessi/versions`. The default is `versions`.


**5. Ingest the tarball into the repository**

Here, we leverage a script from `EESSI/filesystem-layer` that ingests tarballs, but also takes care of regenerating the `.cvmfscatalog` files _and_ updates the `Lmod` cache. To update the `Lmod` cache, the script uses the Lmod installation provided by `software.eessi.io`, which is why we explicitly made this available as one of the steps [when we set up our Stratum 0](#setting-up-your-stratum-0)

Copy link
Copy Markdown
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Suggested change
Here, we leverage a script from `EESSI/filesystem-layer` that ingests tarballs, but also takes care of regenerating the `.cvmfscatalog` files _and_ updates the `Lmod` cache. To update the `Lmod` cache, the script uses the Lmod installation provided by `software.eessi.io`, which is why we explicitly made this available as one of the steps [when we set up our Stratum 0](#setting-up-your-stratum-0)
Here, we leverage [`ingest-tarball.sh`](https://github.com/EESSI/filesystem-layer/blob/main/scripts/ingest-tarball.sh) from `EESSI/filesystem-layer`. The script ingests the tarball, regenerates the `.cvmfscatalog` files, and updates the Lmod cache.
The tarball contains paths relative to the versioned installation directory, such as `2025.06/software/...`. The `--basedir` option determines where that directory is placed relative to the root of the target CVMFS repository. It must therefore match the repository layout configured through `VERSIONS_SUBPATH`.
When updating the Lmod cache, the script uses the selected base directory for the site installation. If the site repository does not provide its own compatibility layer, it automatically falls back to an Lmod installation from `/cvmfs/software.eessi.io/versions`.

Comment on lines +1369 to +1376
echo "Ingesting into CVMFS (${REPO_NAME})..."
if $INGEST_SCRIPT "${REPO_NAME}" "${local_tar}"; then
echo "Ingest succeeded for ${filename}."
else
echo "ERROR: cvmfs_server ingest failed for ${filename}." >&2
# Keep the files for troubleshooting
continue
fi

Copy link
Copy Markdown
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Suggested change
echo "Ingesting into CVMFS (${REPO_NAME})..."
if $INGEST_SCRIPT "${REPO_NAME}" "${local_tar}"; then
echo "Ingest succeeded for ${filename}."
else
echo "ERROR: cvmfs_server ingest failed for ${filename}." >&2
# Keep the files for troubleshooting
continue
fi
echo "Ingesting into CVMFS (${REPO_NAME}) under ${VERSIONS_SUBPATH}..."
if "${INGEST_SCRIPT}" \
--repository "${REPO_NAME}" \
--basedir "${VERSIONS_SUBPATH}" \
"${local_tar}"; then
echo "Ingest succeeded for ${filename}."
else
echo "ERROR: CVMFS ingestion failed for ${filename}." >&2
# Keep the files for troubleshooting
continue
fi


**6. Regenerate the `.cvmfscatalog` files by publishing an empty transaction**

This is already taken care of by the `$INGEST_SCRIPT` in the step above. If you don't want to use that script, you'll have to implement this step yourself.

Copy link
Copy Markdown
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Suggested change
This is already taken care of by the `$INGEST_SCRIPT` in the step above. If you don't want to use that script, you'll have to implement this step yourself.
This is already taken care of by `$INGEST_SCRIPT` in the step above. The patterns in `.cvmfsdirtab` must use the same repository-relative versions path as `VERSIONS_SUBPATH`; otherwise the expected nested catalogs will not be created.


**7. Open a new transaction, update the Lmod cache for your site installs, and publish the transaction**

This is already taken care of by the `$INGEST_SCRIPT` in step 5 above. If you don't want to use that script, you'll have to implement this step yourself.

Copy link
Copy Markdown
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Suggested change
This is already taken care of by the `$INGEST_SCRIPT` in step 5 above. If you don't want to use that script, you'll have to implement this step yourself.
This is already taken care of by `$INGEST_SCRIPT` in step 5. The cache is updated below `${VERSIONS_SUBPATH}/<eessi_version>` in the site repository. The script will use an Lmod installation from the site repository when one is available, and otherwise falls back to `software.eessi.io`.

BUCKET="<bucket_name>"
DOWNLOAD_DIR=/prefix/for/tarball/staging # Some directory to temporarily store tarballs on the Stratum 0
ALLOWED_SIGNERS=/path/to/allowed/signers/file # Optional, needed in step 4
REPO_NAME="<repo_name>"

Copy link
Copy Markdown
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Suggested change
REPO_NAME="<repo_name>"
REPO_NAME="<repo_name>"
# Repository-relative path to the versions directory.
# This must match the versions_subpath used for .cvmfsdirtab.
VERSIONS_SUBPATH="versions"
# For a site prefix such as /cvmfs/name.sitename.tld/eessi, use:
# VERSIONS_SUBPATH="eessi/versions"

Comment on lines +1568 to +1576
# ---- Ingest into CVMFS ----
echo "Ingesting into CVMFS (${REPO_NAME})..."
if $INGEST_SCRIPT "${REPO_NAME}" "${local_tar}"; then
echo "Ingest succeeded for ${filename}."
else
echo "ERROR: cvmfs_server ingest failed for ${filename}." >&2
# Keep the files for troubleshooting
continue
fi

Copy link
Copy Markdown
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Suggested change
# ---- Ingest into CVMFS ----
echo "Ingesting into CVMFS (${REPO_NAME})..."
if $INGEST_SCRIPT "${REPO_NAME}" "${local_tar}"; then
echo "Ingest succeeded for ${filename}."
else
echo "ERROR: cvmfs_server ingest failed for ${filename}." >&2
# Keep the files for troubleshooting
continue
fi
# ---- Ingest into CVMFS ----
echo "Ingesting into CVMFS (${REPO_NAME}) under ${VERSIONS_SUBPATH}..."
if "${INGEST_SCRIPT}" \
--repository "${REPO_NAME}" \
--basedir "${VERSIONS_SUBPATH}" \
"${local_tar}"; then
echo "Ingest succeeded for ${filename}."
else
echo "ERROR: CVMFS ingestion failed for ${filename}." >&2
# Keep the files for troubleshooting
continue
fi

@Neves-P
Neves-P requested a review from bedroge September 18, 2026 13:02
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

Add documentation on how to build on top of EESSI as a site - and include the use of EESSI_SITE_SOFTWARE_PREFIX in that

4 participants