Skip tests relying on quay.io, the Tool Sheds, Dockstore and usegalaxy.eu when they are down

[jmchilton]

Planemo already had one of these — skip_if_zenodo_down, added in 423898f. This applies the
same skip_if_site_down mechanism to the rest of the tests that need a live external service,
mirroring galaxyproject/galaxy#23685 on the Galaxy side.

Motivating failure: a quay.io read timeout failed test_data_manager_docker_mount in CI on an
unrelated branch. Container resolution timed out, so the job ran without the biocontainer,
bowtie2 wasn't there, and the data manager JSON didn't match:

urllib3.exceptions.ReadTimeoutError: HTTPSConnectionPool(host='quay.io', port=443):
    Read timed out. (read timeout=12)
galaxy.tool_util.deps.containers ERROR Could not get container description for tool
    'bowtie2_index_builder_data_manager'
...
galaxy.tool_util.verify INFO: ## files diff ... found diff = 2

New probes

In tests/test_utils.py, beside the existing skip_if_zenodo_down:

skip_if_dockstore_down      = skip_if_site_down("https://dockstore.org/")
skip_if_quay_down           = skip_if_site_down("https://quay.io/")
skip_if_toolshed_down       = skip_if_site_down("https://toolshed.g2.bx.psu.edu/")
skip_if_testtoolshed_down   = skip_if_site_down("https://testtoolshed.g2.bx.psu.edu/")
skip_if_usegalaxy_eu_down   = skip_if_site_down("https://usegalaxy.eu/")

galaxy-util 26.0.1 exports only skip_if_github_down and skip_if_workflowhub_down —
skip_if_quay_down, skip_if_dockstore_down and skip_if_galaxy_depot_down landed in Galaxy
dev with #23685 but aren't in the pinned release yet. So these are defined locally against the
shared skip_if_site_down primitive and can collapse into upstream imports when the pin moves.
The two that do exist upstream are imported from galaxy.util.unittest_utils rather than
re-exported.

Tests decorated

service	tests
quay.io	test_data_manager_docker_mount, test_workflow_test_repository_installation_gxformat2, test_serve_interactivetool — all --biocontainers
Tool Shed	7 tool id/version linting tests in test_cmd_workflow_lint.py; 5 workflow autoupdate tests (test_autoupdate_workflow_from_multiple_tool_sheds also gets skip_if_testtoolshed_down)
Dockstore	test_run_trs_id
github.com	test_ok_http (lints a tool over https from raw.githubusercontent.com), test_git_ls_remote
usegalaxy.eu	test_training_generate_from_wf_command_remote_wf, test_training_init_command_tutorial_remote_wf

Each new decorator sits innermost, below the existing @skip_if_environ(...) marks, so the env
gate is still a collection-time pytest mark and the network probe only runs if the test would
actually execute.

tests/test_trs_id.py is deliberately untouched — it mocks Dockstore with responses, so it
needs no probe.

Verification

• All seven probe URLs return HTTP 200 (is_site_up requires exactly 200, so a URL that never
  returns 200 would skip its tests permanently — worth checking, and they're clean).
• Confirmed the decorator actually skips: pointed one at an unresolvable host and it raised
  SkipTest: Test depends on [...] being up and it appears to be down.
• Ran the decorated tests that don't need a managed Galaxy — 7 tool shed linting tests plus
  test_git_ls_remote and test_ok_http: 9 passed.
• Whole suite still collects: 593 tests, no import errors.
• black / isort / flake8 clean.

Caveat

The probes hit root pages, so they catch a service that is unreachable, not a slow API behind a
working front page. The quay.io failure above was a 12s read timeout on the API while the front
page was presumably fine — so this would not have caught that specific one. It is the same
tradeoff #23685 accepted upstream; a tighter probe would mean per-service API URLs and more
maintenance.

🤖 Generated with Claude Code