myna/spread.yaml at main · canonical/myna · GitHub

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
# Spread project for myna (feature 006, US5) — confined multi-system
# end-to-end tests on clean systems. Decision record:
# specs/006-coverage-quality-gates/spread-decision.md
#
# Run locally (needs KVM):   make spread        (whole suite; primes the VM
#                             image and builds spread on first run)
# Single task:  make spread-<dir under tests/spread>, e.g. spread-adapter-smoke
# Debug a failure (keep the VM):  make spread-adapter-smoke SPREAD_FLAGS=-debug
#
# spread is built from the pinned commit by dev/spread-build.sh (same as CI),
# NOT the snap (no kvm plug), with one local patch: `-cpu host` for the guest
# (qemu64 lacks x86-64-v2; numpy/CTranslate2 wheels refuse to run without it).
#
# The snaps under test are prebuilt and sent with the project:
#   myna-snap/myna_*_amd64.snap              (myna-snap/dev/prepare.sh && snapcraft pack)
#   fake-snap/myna-fake-backend_*_amd64.snap (fake-snap/dev/prepare.sh && snapcraft pack)
#   whisper-snap/myna-whisper_*_amd64.snap   (make snap-whisper)
#   whisper-snap/myna-whisper+model-tiny.comp (same; download-models.sh stages the weights)
#
# confined-e2e uses the fake backend (no model, runs anywhere); adapter-smoke
# uses the real whisper snap to prove a shipped adapter is wired up in both
# emission modes. 4 GB of VM RAM is enough for whisper-tiny on CPU.
project: myna

path: /home/spread/myna

backends:
    qemu:
        memory: 4G
        systems:
            - ubuntu-24.04-64:
                username: ubuntu
                password: ubuntu

# Keep the send small: the suite needs only the two snaps and tests/.
exclude:
    - .git
    - client/target
    - client
    - server/.venv
    - server
    - corpus
    - results
    - .coverage-work
    - .workshop
    - specs
    - docs
    - dev
    - extensions
    - .claude
    - .pi
    - .ruff_cache
    # Local caches and reference checkouts: HF dataset caches, the locally
    # built spread binary + clone, upstream snap sources. Several GB that the
    # tasks never touch - they alone fill the 4 GB guest disk.
    - .cache
    - reference
    - htmlcov
    - myna-bench.pyz
    # whisper-snap is *not* excluded wholesale: adapter-smoke needs the packed
    # snap and the tiny component (161 MB + 76 MB). Its build trees and the
    # larger components are dropped instead - shipping base/small/CUDA would add
    # 2.3 GB to every send for no extra coverage.
    - whisper-snap/parts
    - whisper-snap/stage
    - whisper-snap/prime
    - whisper-snap/stage_packages
    - whisper-snap/components
    - whisper-snap/wheels
    - whisper-snap/myna-whisper+model-base.comp
    - whisper-snap/myna-whisper+model-small.comp
    - whisper-snap/myna-whisper+gpu-nvidia.comp
    - parakeet-snap
    - funasr-snap
    - fake-snap/stage
    - fake-snap/prime
    - fake-snap/parts
    - fake-snap/stage_packages
    - myna-snap/stage
    - myna-snap/prime
    - myna-snap/parts
    - myna-snap/stage_packages

suites:
    tests/spread/:
        summary: Confined end-to-end tests
        # Every task also fails on an AppArmor denial of our snaps, its own
        # restore included (tests/lib/apparmor_denials.py).
        prepare-each: |
            journalctl -n 0 --show-cursor --no-pager | sed -n 's/^-- cursor: //p' > /root/apparmor.cursor
        restore-each: |
            journalctl --after-cursor "$(cat /root/apparmor.cursor)" --no-pager \
                | python3 -I "$SPREAD_PATH/tests/lib/apparmor_denials.py"