- Shipped
- August 9, 2026 at 2:26 AM UTC
- Author
- Kamo
- Commit
- 04fe24f
Three builds died today on "Tag mismatch" pulling a jar — and the mirror added an hour ago did not help, because kbservice, mcpgatewayservice and ragservice hit the identical error against the Google mirror itself. So this was never Maven Central rate-limiting. It is k1m1 eno49: a measured 15-35% of transfers over ~20MB corrupt on that segment (reference: the 2026-08-04 localisation — clean over eno50 p2p, clean workstation-to-k3m1, corrupt only through that switch port). The same fault took out the puppeteer download in kamo-login with an SSL bad-record-mac. TLS notices the corruption and aborts, so the failure is loud rather than a bad jar being cached, and a retry usually lands. Three attempts around the shared-lib install turns a coin-flip into a rare failure. SecurityService already did this, which is why it kept passing while its neighbours did not. This is a workaround for faulty hardware. The cable or switch port for eno49 still needs replacing; until then every large download on this runner is a dice roll.