Compare commits

..

2 Commits

Author SHA1 Message Date
Niklas Ye 43f004499b Set the chart's placeholder version to 0.20.1
CI / chart (push) Successful in 1s
CI / security (push) Successful in 13s
CI / test (push) Successful in 2m30s
Release / test (push) Successful in 4s
Release / chart (push) Successful in 2s
Release / binaries (push) Successful in 15s
Release / image (push) Successful in 55s
Release / scan-image (push) Successful in 2s
Cosmetic: make helm-package sets the published version and appVersion
from the tag, so these two fields decide nothing (see the comment
above them). Kept in step anyway, same as e77f04b and 429d5fd, so a
tree heading for v0.20.1 doesn't say 0.20.0.

Claude-Session: https://claude.ai/code/session_01RHPj4ggeFdEjKKfm4SHbD7
2026-09-22 22:42:13 +02:00
Niklas Ye 559be6de6e Retry the first database ping instead of dying on it
Every start crashed once or twice before going healthy: kube-router
enforces this namespace's NetworkPolicy per-node, reacting to the new
pod's creation event, and the app's first connection attempt can
reach Postgres's node before that node's allow-set has been updated
with the new pod's IP. The result is "connection refused" -- an
active reject, not a timeout, which is how it was told apart from
Postgres itself not being ready (it had been up for two days in the
run that was diagnosed).

That race resolves within several seconds in practice, so Open now
retries the ping up to five times, two seconds apart, logging each
failure, before giving up with the same wrapped error as before.
Nothing else about Open's behaviour changed: a genuinely absent
database still fails, just after ~8s instead of immediately.

Claude-Session: https://claude.ai/code/session_01RHPj4ggeFdEjKKfm4SHbD7
2026-09-22 22:42:05 +02:00
2 changed files with 28 additions and 6 deletions
+2 -2
View File
@@ -15,5 +15,5 @@ type: application
# appVersion and image.tag in values.yaml no longer agree, and that is not an oversight:
# image.tag stays "latest", which is what a local install actually pulls. appVersion is
# metadata and drives nothing.
version: 0.20.0
appVersion: "v0.20.0"
version: 0.20.1
appVersion: "v0.20.1"
+24 -2
View File
@@ -5,6 +5,7 @@ import (
"embed"
"fmt"
"io/fs"
"log"
"sort"
"strings"
"time"
@@ -15,6 +16,19 @@ import (
//go:embed migrations
var migrationsFS embed.FS
// pingAttempts and pingRetryDelay bound the retry on the first connection.
// This pod's own IP can reach the Postgres pod's node before that node's
// NetworkPolicy enforcement (kube-router, reacting to the pod's creation
// event) has added it to the allowed-source set, which fails the ping with
// "connection refused" rather than a timeout. That race resolves within
// several seconds in practice; five attempts two seconds apart give it
// comfortable room without turning a genuinely absent database into a long
// hang.
const (
pingAttempts = 5
pingRetryDelay = 2 * time.Second
)
// Open connects to Postgres. dsn is a libpq connection string or URL, e.g.
// postgres://terdut:secret@localhost:5432/terdut?sslmode=disable.
//
@@ -33,11 +47,19 @@ func Open(dsn string) (*sql.DB, error) {
db.SetMaxOpenConns(10)
db.SetMaxIdleConns(5)
db.SetConnMaxLifetime(time.Hour)
if err := db.Ping(); err != nil {
for attempt := 1; ; attempt++ {
err = db.Ping()
if err == nil {
return db, nil
}
if attempt == pingAttempts {
db.Close()
return nil, fmt.Errorf("ping: %w", err)
}
return db, nil
log.Printf("open db: ping attempt %d/%d failed, retrying in %s: %v", attempt, pingAttempts, pingRetryDelay, err)
time.Sleep(pingRetryDelay)
}
}
// Migrate applies every embedded migration that has not been applied yet, in