Repository navigation
Commit 3c024d1
committed
revert(ci): drop the kache experiment; restore rust-cache baseline
Measured on the pool (2026-08-13): kache-only (no rust-cache) ran the rust
job in 30.8 min vs the 27 min rust-cache baseline; kache + rust-cache was
38.1 min. The job is test-dominated (~20 min of ~27), so a compile cache
cannot win: kache removes the rust-cache restore but adds setup/post
overhead (binary download, daemon, warm prefetch, sync push, report).
R2 warm hits covered the compile phase and the result was still slower.
Revert to the proven configuration: Swatinem/rust-cache on the rust job,
no kache-action anywhere. The real fixes from this session stand: the
5 GiB VM memory ceiling (host OOM class) and the deaf-runner fixes
(bridge resilience + server liveness sweep, shipped in v0.30.2).
R2 bucket/secrets stay in place if a compile-heavy workload ever
justifies revisiting this.1 parent 405a391 commit 3c024d1
2 files changed
Lines changed: 2 additions & 36 deletions
| Original file line number | Diff line number | Diff line change | |
|---|---|---|---|
| |||
78 | 78 | | |
79 | 79 | | |
80 | 80 | | |
81 | | - | |
82 | | - | |
83 | | - | |
84 | | - | |
85 | | - | |
86 | | - | |
87 | | - | |
88 | | - | |
89 | | - | |
90 | | - | |
91 | | - | |
92 | | - | |
93 | | - | |
94 | | - | |
| 81 | + | |
| 82 | + | |
95 | 83 | | |
96 | 84 | | |
97 | 85 | | |
| |||
118 | 106 | | |
119 | 107 | | |
120 | 108 | | |
121 | | - | |
122 | | - | |
123 | | - | |
124 | | - | |
125 | | - | |
126 | | - | |
127 | | - | |
128 | | - | |
129 | | - | |
130 | | - | |
131 | 109 | | |
132 | 110 | | |
133 | 111 | | |
| |||
| Original file line number | Diff line number | Diff line change | |
|---|---|---|---|
| |||
143 | 143 | | |
144 | 144 | | |
145 | 145 | | |
146 | | - | |
147 | | - | |
148 | | - | |
149 | | - | |
150 | | - | |
151 | | - | |
152 | | - | |
153 | | - | |
154 | | - | |
155 | | - | |
156 | | - | |
157 | | - | |
158 | 146 | | |
159 | 147 | | |
160 | 148 | | |
| |||
0 commit comments