[Backport perf-v17] fix(integration-tests): stop pinning artifact URLs that get garbage collected

[scylladbbot]

Fixes the three remaining integration test failures from PR-15647 build 1 (unrelated to that PR — a renovate gemini bump). All three are pinned artifact URLs that have since been garbage collected or retired upstream. Companion to #15670, which covers the Secrets Manager failures.

What was failing

Test	Error
test_config_get_version_based_on_conf::test_unified_package	FileNotFoundError: Unified package not found or failed to download: .../2023-11-13T03:04:27Z/scylla-unified-5.5.0~dev-...tar.gz
test_external_backtrace_service::test_decode_via_external_service_returns_symbols	404 ... for url: https://backtrace.scylladb.com/api/backtrace
test_version_utils::test_get_specific_tag_of_docker_image[scylladb/scylla-enterprise-nightly]	404 ... /unstable/scylla-enterprise/enterprise/relocatable/latest/00-Build.txt

Fixes

Unified package. The test pinned a master relocatable from November 2023. Unlike its sibling tests, which only run config validation, this one calls get_version_based_on_conf(), which downloads and unpacks the package to read SCYLLA-VERSION-FILE — so the URL has to actually exist, and unstable relocatables are collected within months. Now resolved through the existing latest_unified_package() helper, with the expected version derived from the returned filename.

Backtrace service. The pinned build id (527bc254…, 2026.1.0~dev) is still indexed, but the unstripped package it needs is gone, so the API answers:

{"detail":"Unstripped package does not exist (https://downloads.scylladb.com/unstable/scylla/master/
relocatable/2026-01-22T20:32:01Z/scylla-unstripped-2026.1.0~dev-...tar.gz) for 527bc254..."}

Now resolved from unstable/scylla/master/relocatable/latest/00-Build.txt. Newer builds exist in S3 but the service does not index them until they are promoted (checked: today's 14:11 and 16:36 builds return Could not find any build with BuildID, while latest/ resolves and symbolizes), so latest/ is the freshest reliably-known id. If it happens not to be indexed yet the test skips rather than reporting a failure that says nothing about SCT.

Enterprise nightly. get_specific_tag_of_docker_image resolved scylladb/scylla-enterprise-nightly under the enterprise rolling branch. That branch is gone from downloads.scylladb.com — unstable/scylla-enterprise/ now holds only enterprise-2024.1/, which is still building (last build June 2026) and still publishes exactly those images:

docker-image-name-x86_64: scylla-enterprise-nightly:2024.1.22-0.20260125.edbbdaf63453-x86_64

So the lookup points there. This is a production fix, not just a test fix — sct_config calls this function to resolve scylla_docker_image for enterprise versions and it was raising for every caller.

Testing

unit_tests/integration/test_external_backtrace_service.py
unit_tests/integration/test_version_utils.py                            14 passed
unit_tests/integration/test_config_get_version_based_on_conf.py -k unified
                                                                 8 passed, 25 deselected
unit_tests/unit/test_version_utils.py unit_tests/unit/test_config.py   417 passed

• (cherry picked from commit c481dd1)
• (cherry picked from commit 0716d88)
• (cherry picked from commit 4698ab6)

Parent PR: #15671

[scylladbbot]

@fruch - This PR has conflicts, therefore it was moved to draft
Please resolve them and mark this PR as ready for review by removing the conflicts label

[fruch]

Resolved the cherry-pick conflicts and dropped one commit on purpose — please don't restore it.

Dropped: fix(integration-tests): resolve a live build id for the backtrace service (master 4698ab66d8).

On master that commit edits unit_tests/integration/test_external_backtrace_service.py; the file does not exist here, so the cherry-pick turned it into a whole-new-file add. That file cannot work on branch-perf-v17:

• BaseNode._decode_via_external_service — the method under test — does not exist on this branch. perf-v17 still decodes backtraces locally (copy_scylla_debug_info + decode_raw_backtrace); the external-service decode path was never backported.
• sdcm/utils/session.py (create_retry_session) does not exist here either, so the module fails at import: ModuleNotFoundError: No module named 'sdcm.utils.session'.
• The remaining imports point at master's layout (unit_tests.lib.dummy_remote, unit_tests.lib.fake_cluster); on this branch they live at unit_tests/dummy_remote.py and unit_tests/nemesis/fake_cluster.py, and there is no unit_tests/integration/ package at all.

Backporting the test would mean backporting the external-decode feature too, which is well outside the scope of "stop pinning artifact URLs". So it is dropped, and the branch no longer adds unit_tests/integration/.

Kept — both verified against this branch's code and layout:

1. fix(version_utils): resolve enterprise-nightly tags from enterprise-2024.1 — applies cleanly. Verified unstable/scylla-enterprise/enterprise/relocatable/latest/00-Build.txt → 404 and .../enterprise-2024.1/... → 200 with a docker-image-name key; test_get_specific_tag_of_docker_image[scylladb/scylla-enterprise-nightly] passes.
2. fix(integration-tests): resolve the unified package URL at run time — landed in this branch's real file, unit_tests/test_config_get_version_based_on_conf.py (rename detection followed master's move into unit_tests/integration/, so no duplicate copy was created). latest_unified_package() already exists in sdcm/utils/version_utils.py here, so no helper had to come along. test_unified_package passes, and the URL it used to pin still 404s.

Checks run locally on the pushed tree: ruff check --preview and ruff format --check clean on both files, full pre-commit run --files ... green, the two tests above pass, and no conflict markers remain in any file of any commit (every commit's blobs checked with no extension exclusions).

Pre-existing failures on this branch, unrelated to this backport — please don't read them as regressions:

• test_get_specific_tag_of_docker_image[scylladb/scylla-nightly] and test_docker[latest] → KeyError: 'docker-image-name'. unstable/scylla/master/.../00-Build.txt now publishes only docker-image-name-x86_64; master's get_specific_tag_of_docker_image handles that with an arch-aware lookup that was never backported here.
• test_unified_package_aws_sets_ubuntu_user → FileNotFoundError: unit_tests/test_configs/minimal_test_case.yaml (that fixture config does not exist on this branch).
• test_images[azure-master], test_images[azure-branch] → Azure image lookup finds nothing in eastus.

[fruch]

Integration test results (build #2) — this PR's fix is confirmed working by CI

continuous-integration/jenkins/pr-merge = fail with "The build of this commit was aborted". This is not a defect in this backport:

• jenkins/precommit pass, jenkins/unittests pass, jenkins/lint_test_cases pass
• The integration stage hit its 45-minute timeout: Cancelling nested steps due to timeout → Timeout has been exceeded → No test report files were found. Same abort as [Backport perf-v17] fix(integration-tests): align AWS keystore tests with the Secrets Manager backend #15680 on this branch and [Backport 2025.1] fix(integration-tests): align AWS keystore tests with the Secrets Manager backend #15690 on branch-2025.1.

Direct evidence the fix works, by diffing this run against #15680's run on the same branch and same integration suite (the two differ only by this PR's change):

Test	#15680's run	this PR's run
test_config_get_version_based_on_conf.py::test_unified_package	FAILED	passes ✅

That is the test this backport targets, and it flipped green. It is no longer in the failure list at all.

The remaining failures are all pre-existing on branch-perf-v17 and untouched by this PR — identical to #15680's set apart from the line above:

unit_tests/test_cassandra_stress_thread.py  (9 cases)
unit_tests/test_cluster.py                  (2 cases)
unit_tests/test_config.py                   (3 cases)
unit_tests/test_config_get_version_based_on_conf.py::test_docker[latest]
unit_tests/test_config_get_version_based_on_conf.py::test_images[azure-branch|azure-master]
unit_tests/test_config_get_version_based_on_conf.py::test_unified_package_aws_sets_ubuntu_user

Two of those have verified causes worth recording:

1. test_unified_package_aws_sets_ubuntu_user — unit_tests/test_configs/ does not exist on branch-perf-v17 (git ls-tree upstream/branch-perf-v17 unit_tests/test_configs/ → empty; present on master, branch-2026.1, branch-2026.2, branch-2026.3). Tests referencing unit_tests/test_configs/minimal_test_case.yaml fail at fixture setup. branch-2025.1 has the identical gap.
2. test_docker[latest] — 00-Build.txt for master now publishes only docker-image-name-x86_64, not the legacy docker-image-name key, and this branch lacks master's arch-aware lookup in get_specific_tag_of_docker_image. That is a separate backport candidate, not part of fix(integration-tests): stop pinning artifact URLs that get garbage collected #15671.

Consequence for reviewers: the test-integration stage cannot go green on branch-perf-v17 today for reasons unrelated to this PR, but this PR's own target test is demonstrably fixed. Remediation the branch needs separately: backport/adapt unit_tests/test_configs/, backport the arch-aware docker-image-name lookup, and investigate the 45-minute integration timeout that prevents any JUnit result from being published.