Files
Chris Lu 0973634fd4 telemetry: keep only clusters that store at least 10 GiB (#11138)
* telemetry: tidy the server module after the protobuf bump

Claude-Session: https://claude.ai/code/session_01VGiphDxpsKMwu9XpUFhybQ

* telemetry: keep only clusters that store at least 10 GiB

Fresh weed server runs, CI jobs and throwaway containers each mint their
own cluster id. They came in at tens of thousands a day, were most of
the counted clusters and held almost none of the bytes, and the state
file and the metrics page grew with every one of them. Reports under
the floor are counted and dropped, and a state file written before the
floor sheds them on the first restart.

Claude-Session: https://claude.ai/code/session_01VGiphDxpsKMwu9XpUFhybQ

* master: report telemetry only once the cluster stores 10 GiB

A throwaway cluster no longer registers itself with its first report a
minute after start; a real one begins reporting at the first daily tick
after it crosses the floor.

Claude-Session: https://claude.ai/code/session_01VGiphDxpsKMwu9XpUFhybQ
2026-09-03 12:36:00 -07:00

116 lines
3.0 KiB
Go

package storage
import (
"encoding/json"
"fmt"
"os"
"path/filepath"
"github.com/seaweedfs/seaweedfs/telemetry/proto"
)
// persistedState is the on-disk snapshot of the in-memory instance map.
type persistedState struct {
Instances map[string]*telemetryData `json:"instances"`
Histories map[string][]HistorySample `json:"histories,omitempty"`
}
// LoadState restores the instance map and Prometheus gauges from a state file
// written by SaveStateIfDirty. A missing file is not an error. Original
// ReceivedAt timestamps are preserved so cleanup and the active-cluster
// windows stay correct across restarts. Clusters under proto.MinDiskBytes are
// dropped, so state written before the floor sheds them on the first restart.
func (s *PrometheusStorage) LoadState(path string) (int, error) {
b, err := os.ReadFile(path)
if err != nil {
if os.IsNotExist(err) {
return 0, nil
}
return 0, err
}
var state persistedState
if err := json.Unmarshal(b, &state); err != nil {
return 0, fmt.Errorf("parse %s: %w", path, err)
}
s.mu.Lock()
defer s.mu.Unlock()
loaded := 0
for id, instance := range state.Instances {
if instance == nil || instance.TelemetryData == nil || instance.TelemetryData.TopologyId == "" {
continue
}
if instance.TelemetryData.TotalDiskBytes < proto.MinDiskBytes {
s.dirty = true // so the next save sheds it
continue
}
s.instances[id] = instance
s.setClusterMetrics(instance.TelemetryData)
loaded++
}
for id, h := range state.Histories {
instance, ok := s.instances[id]
if !ok || len(h) == 0 {
continue
}
// State written before versions were recorded carries none on its
// samples. The newest sample is the report the instance record itself
// came from, so that day's version is known and the version series can
// start there rather than a day after the upgrade.
if newest := len(h) - 1; h[newest].Version == "" {
h[newest].Version = instance.TelemetryData.Version
}
s.histories[id] = h
}
s.updateStats()
return loaded, nil
}
// SaveStateIfDirty writes the instance map to path if it changed since the
// last successful save. The write is atomic (temp file + rename).
func (s *PrometheusStorage) SaveStateIfDirty(path string) error {
s.mu.Lock()
if !s.dirty {
s.mu.Unlock()
return nil
}
b, err := json.Marshal(&persistedState{Instances: s.instances, Histories: s.histories})
if err != nil {
s.mu.Unlock()
return err
}
s.dirty = false
s.mu.Unlock()
if err := s.writeAtomically(path, b); err != nil {
s.mu.Lock()
s.dirty = true
s.mu.Unlock()
return err
}
return nil
}
func (s *PrometheusStorage) writeAtomically(path string, b []byte) error {
dir := filepath.Dir(path)
if err := os.MkdirAll(dir, 0755); err != nil {
return err
}
tmp, err := os.CreateTemp(dir, filepath.Base(path)+".tmp-*")
if err != nil {
return err
}
if _, err := tmp.Write(b); err != nil {
tmp.Close()
os.Remove(tmp.Name())
return err
}
if err := tmp.Close(); err != nil {
os.Remove(tmp.Name())
return err
}
return os.Rename(tmp.Name(), path)
}