diff options
| author | Alex Chan <alexc@tailscale.com> | 2025-10-21 12:45:03 +0100 |
|---|---|---|
| committer | Alex Chan <alex@alexwlchan.net> | 2025-10-24 15:29:07 +0100 |
| commit | d47c697748ec2cf0d3ca663811b094ec617529cd (patch) | |
| tree | 12a5c7edee0dda942a0da9086af3fa2bd39404e7 /util/execqueue/execqueue_test.go | |
| parent | 8576a802caabffd5c5e94d614acc8dc954f0a443 (diff) | |
| download | tailscale-d47c697748ec2cf0d3ca663811b094ec617529cd.tar.xz tailscale-d47c697748ec2cf0d3ca663811b094ec617529cd.zip | |
ipn/ipnlocal: skip TKA bootstrap request if Tailnet Lock is unavailable
If you run tailscaled without passing a `--statedir`, Tailnet Lock is
unavailable -- we don't have a folder to store the AUMs in.
This causes a lot of unnecessary requests to bootstrap TKA, because
every time the node receives a NetMap with some TKA state, it tries to
bootstrap, fetches the bootstrap TKA state from the control plane, then
fails with the error:
TKA sync error: bootstrap: network-lock is not supported in this
configuration, try setting --statedir
We can't prevent the error, but we can skip the control plane request
that immediately gets dropped on the floor.
In local testing, a new node joining a tailnet caused *three* control
plane requests which were unused.
Updates tailscale/corp#19441
Signed-off-by: Alex Chan <alexc@tailscale.com>
Diffstat (limited to 'util/execqueue/execqueue_test.go')
0 files changed, 0 insertions, 0 deletions