Repository navigation
Commit ad35afa
committed
fix(e2e): disable guest fsync on ephemeral CI VM disks
The e2e QEMU guests boot with three raw virtio -drive lines that carry
no cache= option, so QEMU defaults to cache=writeback and honors every
guest fsync. The management-cluster Talos etcd fsyncs on each Raft
commit, while containerd unpacks dozens of images and CDI imports
multi-GB VM disks onto the same host-backed raw files during the
VM-heavy suites. The fsync storm serializes etcd's writes into
multi-second stalls, the mgmt apiserver times out lease PUTs, and many
operators (capk, etcd-operator, fdb-operator, cilium-operator, kamaji)
lose leadership at once — failing whichever chainsaw suite is
mid-flight, a different one each run.
Add cache=unsafe to each guest drive so guest flushes become no-ops
against the host page cache, removing the fsync-gated etcd stalls. This
is safe only because these CI VMs are ephemeral and destroyed every
run; never use it for real data.
Assisted-By: Claude <[email protected]>
Signed-off-by: Andrei Kvapil <[email protected]>1 parent d598e67 commit ad35afa
1 file changed
Lines changed: 3 additions & 3 deletions
| Original file line number | Diff line number | Diff line change | |
|---|---|---|---|
| |||
100 | 100 | | |
101 | 101 | | |
102 | 102 | | |
103 | | - | |
104 | | - | |
105 | | - | |
| 103 | + | |
| 104 | + | |
| 105 | + | |
106 | 106 | | |
107 | 107 | | |
108 | 108 | | |
| |||
0 commit comments