prisma-risk / tsoracle / 33047928024
96%

DEFAULT BRANCH: main
Ran
Jobs 1
Files 97
Run time 1min
Badge
Embed ▾
README BADGES
x

If you need to use a raster PNG badge, change the '.svg' to '.png' in the link

Markdown

Textile

RDoc

HTML

Rst
coverage: 96.433% (-0.02%) from 96.448%
33047928024

push

github

web-flow 
fix(deploy): stop liveness owning the boot window (#691)

Both workloads shipped a readiness and a liveness probe but no startupProbe, so liveness owned the boot window: with initialDelaySeconds 10, periodSeconds 10, failureThreshold 3, the kubelet killed the container at t≈40s. A container killed mid-boot is retried with the same deadline, never a larger one, so a node whose store replay or peer catch-up runs long crash-loops rather than starting slowly — and on an HA driver every replica is in that window at once.

Adds a startupProbe with a 120-second budget to the file-driver Deployment and the HA StatefulSet alike, and raises liveness.initialDelaySeconds to 120 so liveness covers the budget independently (kubernetes/kubernetes#105308). Probe timings move into a probes block shared by both workloads.

Records the gap this cannot close: all three probes are the same tcpSocket check, because tsoracle exposes no health surface. Readiness therefore reports only that the port accepts connections and cannot distinguish a node that can mint timestamps from one that is up but has lost quorum, so an HA replica that cannot serve stays in the client Service. Closing that needs a real readiness signal, and what "ready" means for the oracle is a design decision rather than a chart one.

Signed-off-by: Sebastian Thiebaud <sebastian@prismarisk.com>

18354 of 19033 relevant lines covered (96.43%)

298009.81 hits per line

Coverage Regressions

Lines Coverage ∆ File
2
80.46
 -0.57% crates/tsoracle-paxos-toolkit/src/lifecycle/mod.rs
1
95.29
 -0.06% crates/tsoracle-client/src/retry.rs
Jobs
ID Job ID Ran Files Coverage
1 33047928024.1 97
96.43
 GitHub Action Run
Source Files on build 33047928024