[fstagni]

CI fix: install conda explicitly in the PilotWrapper workflow

Summary

The PilotWrapper jobs have been failing on every PR since 2026-11-09 (runner image ubuntu26/20261101) with:

prepare environment: line 1: conda: command not found

The workflow assumes conda is preinstalled and on the PATH of ubuntu-latest runners — the new image no longer provides it. This PR provisions it explicitly:

       - name: Fail-fast for outdated pipelines
         run: .github/workflows/fail-fast.sh
+
+      # conda is no longer preinstalled (nor on the PATH) of the ubuntu-latest
+      # runner images; install it explicitly since the next steps rely on it
+      - name: Install conda
+        uses: conda-incubator/setup-miniconda@v3
+        with:
+          auto-update-conda: false
+          activate-environment: ""
 
       - name: prepare environment
         run: |
           conda config --set add_pip_as_python_dependency false

Two other workflows had the same problem (raw conda + the Miniconda PATH hack, both gone from the new image) and get the same treatment: cvmfs.yml (Deployment_test) and deployment.yml (deploy_CVMFS). Both are workflow_dispatch only, so their breakage was invisible on PRs.

Everything else — the conda config/conda create invocations, and the test itself — is unchanged.

Evidence

• Failure timeline: pilotWrapper.yml runs pass up to 2026-11-06 (including this org's other PR branches), fail from 2026-11-09 regardless of branch — e.g. feat/add-custom-param-to-ps (runs 4259707... / 08:14 and 09:22 UTC) and feat/hs23-cpu-normalization (also retried once, same conda: command not found).
• The runner image change is visible in the job logs: "Image Release: ubuntu26/20261101.186".
• This PR's own PilotWrapper jobs will exercise the fix directly.

Merge Danger

Door: two-way — one workflow step added; if setup-miniconda misbehaves the jobs fail exactly as they do today.

Blast Radius: CI only; no product code.

[aldbr]

Not sure this is expected: https://github.com/DIRACGrid/DIRAC/actions/runs/42600472020/job/129103370911?pr=8876#step:7:22

And https://github.com/DIRACGrid/DIRAC/actions/runs/42600472020/job/129103370889?pr=8876#step:6:48

[aldbr]

Aren't you going to have the same issue with cvmfs.yml?

DIRAC/.github/workflows/cvmfs.yml

Lines 8 to 17 in d46e17f

	runs-on: "ubuntu-latest"
	steps:
	- name: Install system packages
	run: |
	echo "===== Install fetch-crl as a system package ====="
	sudo apt-get update
	sudo apt-get install -y fetch-crl
	- name: prepare environment
	run: |
	conda create -c conda-forge -n cvmfs-env ca-policy-lcg openssl==3.0.7 gct

[fstagni]

All checks are green (23 passing; the two skipped jobs are intentional). Note that the approval was auto-dismissed by dismiss_stale_reviews after the last two pushes (the 3.6.9 matrix pin replacing 3.6.8 — which only existed in the defaults channel — and the same conda treatment for cvmfs.yml/deployment.yml), so the PR now shows BLOCKED pending a re-approval. Once merged, #8870 and #8874 will be rebased/updated onto the fix.