Port install_agfhc to the orch fixture; drop nfs_install [AIMVT-170]#181
Merged
Conversation
Mechanical translation of phdl/shdl -> orch per orch fixture pattern (parent: AIMVT-2). Deletes the local phdl/shdl fixtures (orch lives in cvs/tests/conftest.py per AIMVT-172). Also removes the nfs_install knob from this suite. Precedent: install_rvs.py (merged via PR #174) already silently ignores the field. The head-only optimization saved only a cp+tar extraction step -- sudo ./install was already always fleet-wide (legacy L172). After this PR, install runs unconditionally on every node via orch.exec. Behavior change: package_tar_ball must now exist at the same local path on every node. Operators who previously relied on nfs_install=True with the tarball only on a shared mount must pre-stage the tarball per-node. Out of scope: the rvs.nfs_install and transferbench.nfs_install keys in mi300_health_config.json remain (covered by separate PRs).
Splits `cd X; cp/tar/install` shell-string chains into separate orch.exec calls, each invoking a single binary with absolute paths. This makes the test's command flow legible at the call site and works correctly under ContainerOrchestrator's docker-exec transport, which does not spawn a shell to interpret `cd`, `;`, `&&` and friends. The single AGFHC `./install` call still requires cwd = install_dir (installer reads sibling files via relative paths), so that one call is wrapped in an explicit `sudo bash -c 'cd X && ./install'` to make the cwd dependency visible at the call site rather than smuggling it in via a shell chain. Also passes `--rocm-tar` to AGFHC's installer: this bypasses apt-based dependency resolution (which fails on TheRock-rocm systems because the libs aren't tracked by dpkg) in favour of dpkg-deb direct extraction. Safe on apt-rocm hosts too: dpkg-deb extracts the bundled debs into /, which is correct either way. Validated on a06u43 (single MI300X node, ROCm 7.13.0 in cvs-harness/ rocm-dev:latest container, externally-managed via launch:false): 1 passed in 26.13s. Container /opt/amd/agfhc/agfhc now exists post-run.
amd-droy
pushed a commit
that referenced
this pull request
Jun 25, 2026
…181) * [AIMVT-170] Port install_agfhc.py to orch fixture and drop nfs_install Mechanical translation of phdl/shdl -> orch per orch fixture pattern (parent: AIMVT-2). Deletes the local phdl/shdl fixtures (orch lives in cvs/tests/conftest.py per AIMVT-172). Also removes the nfs_install knob from this suite. Precedent: install_rvs.py (merged via PR #174) already silently ignores the field. The head-only optimization saved only a cp+tar extraction step -- sudo ./install was already always fleet-wide (legacy L172). After this PR, install runs unconditionally on every node via orch.exec. Behavior change: package_tar_ball must now exist at the same local path on every node. Operators who previously relied on nfs_install=True with the tarball only on a shared mount must pre-stage the tarball per-node. Out of scope: the rvs.nfs_install and transferbench.nfs_install keys in mi300_health_config.json remain (covered by separate PRs). * [AIMVT-170] install_agfhc: one logical command per orch.exec; --rocm-tar Splits `cd X; cp/tar/install` shell-string chains into separate orch.exec calls, each invoking a single binary with absolute paths. This makes the test's command flow legible at the call site and works correctly under ContainerOrchestrator's docker-exec transport, which does not spawn a shell to interpret `cd`, `;`, `&&` and friends. The single AGFHC `./install` call still requires cwd = install_dir (installer reads sibling files via relative paths), so that one call is wrapped in an explicit `sudo bash -c 'cd X && ./install'` to make the cwd dependency visible at the call site rather than smuggling it in via a shell chain. Also passes `--rocm-tar` to AGFHC's installer: this bypasses apt-based dependency resolution (which fails on TheRock-rocm systems because the libs aren't tracked by dpkg) in favour of dpkg-deb direct extraction. Safe on apt-rocm hosts too: dpkg-deb extracts the bundled debs into /, which is correct either way. Validated on a06u43 (single MI300X node, ROCm 7.13.0 in cvs-harness/ rocm-dev:latest container, externally-managed via launch:false): 1 passed in 26.13s. Container /opt/amd/agfhc/agfhc now exists post-run.
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Tracking: AIMVT-170.
Changes
cvs/tests/health/install/install_agfhc.py:from cvs.lib.parallel_ssh_lib import *and the localphdl/shdlfixtures. The test signature becomes
test_install_agfhc(orch, config_dict).nfs_installhead-vs-fleet conditional. Install runsunconditionally on every node via
orch.exec(matches the mergedinstall_rvs.pybehavior). Thepackage_tar_ballmust exist at the samelocal path on every node.
cd X; cp ... && tar -xvf ...) intoseparate single-binary
orch.execcalls. The single./installinvocationthat genuinely needs
cwd = install_dir(AGFHC's installer reads siblingfiles via relative paths) is wrapped in an explicit
sudo bash -c 'cd X && ./install'so the cwd dependency is visible at thecall site instead of relying on an implicit shell. Each call now invokes one
binary, which is what
ContainerOrchestrator's docker-exec transportrequires.
--rocm-tarto AGFHC's installer so it bypasses apt-based dependencyresolution in favour of
dpkg-debextraction. Required on TheRock-rocmhosts (libs not tracked by dpkg); no-op on apt-rocm hosts.
cvs/input/config_file/health/mi300_health_config.json:"nfs_install": "True"from theagfhcblock.docs/reference/configuration-files/health.rst:"nfs_install"entry from the inline JSON dropdown andthe
nfs_installrow from the AGFHC parameter table.