// reports the calls, the time per call, the calls at once, the cached share,
// the quota, the retries, and the line sizes. BUILDS, from 1 to maxBuilds,
// limits the builds that call the model at once in this run only. The core
-// still queues up to maxBuilds builds.
-func probe(ctx context.Context, args []string, out io.Writer) error {
+// still queues up to maxBuilds builds. The run holds the lock of the token
+// file, so it does not run while the daemon runs.
+func probe(ctx context.Context, args []string, out io.Writer) (err error) {
if len(args) != 2 && len(args) != 3 {
return errors.New("usage: onechat probe LOG COUNT [BUILDS]")
}
return err
}
+ tokens, err := lockTokens(dir)
+ if err != nil {
+ return err
+ }
+
+ defer func() {
+ err = errors.Join(err, tokens.Close())
+ }()
+
address, auth, err := modelEndpoint(config, dir)
if err != nil {
return err
}
usagePath := filepath.Join(root, "usage.jsonl")
- client := newClient(address, auth, "onechat-"+probeName, usagePath)
+ client := newModelClient(address, auth, "onechat-"+probeName, usagePath)
client.LimitCalls(probeCallLimit)
summarizer := slotSummarizer{
- inner: newCompactor(client, probeName, config.Compactor),
+ inner: newCompactor(client, probePrompt, config.Compactor),
slots: make(chan struct{}, builds),
}
// by the rules of probeStop.
type probeRun struct {
memory *Memory
- client *Client
+ client *ModelClient
usagePath string
}
return s.inner.Summarize(ctx, task)
}
+// probePrompt returns the system prompt of a probe run: the prompt of the agent
+// probeName, without instructions.
+func probePrompt() (string, error) {
+ return prompt(probeName, ""), nil
+}
+
// probeEntry is one message of a probe log.
type probeEntry struct {
Kind string `json:"kind"`
// probeStop returns why a probe run must stop: a failed call, or a quota window
// whose used percent grew by more than probeQuotaLimit points since the first
// call. It returns "" when the run can go on.
-func probeStop(records []usageRecord) string {
+func probeStop(records []ModelUsageRecord) string {
for _, record := range records {
if record.Error != "" {
return "call failed: " + record.Error
// readUsageFile reads the records of usage.jsonl at path, in the order of the
// start of their calls. The file has them in the order of the end of their
// calls, which differs when calls run at once.
-func readUsageFile(path string) (records []usageRecord, err error) {
+func readUsageFile(path string) (records []ModelUsageRecord, err error) {
file, err := os.Open(path)
if errors.Is(err, os.ErrNotExist) {
return nil, nil
scanner := bufio.NewScanner(file)
scanner.Buffer(make([]byte, 64*1024), probeLineBytes)
for scanner.Scan() {
- var record usageRecord
+ var record ModelUsageRecord
err = json.Unmarshal(scanner.Bytes(), &record)
if err != nil {
return nil, fmt.Errorf("decode usage: %w", err)
return nil, fmt.Errorf("read usage file: %w", err)
}
- slices.SortStableFunc(records, func(a, b usageRecord) int {
+ slices.SortStableFunc(records, func(a, b ModelUsageRecord) int {
return callStart(a).Compare(callStart(b))
})
// callStart returns the start of the call of record, or the zero time when its
// date is not valid.
-func callStart(record usageRecord) time.Time {
+func callStart(record ModelUsageRecord) time.Time {
start, err := time.Parse(time.RFC3339Nano, record.Date)
if err != nil {
return time.Time{}
// usedPercents returns the used percent of each quota window of record, by
// header name.
-func usedPercents(record usageRecord) map[string]float64 {
+func usedPercents(record ModelUsageRecord) map[string]float64 {
used := make(map[string]float64)
for name, value := range record.Quota {
if !strings.HasSuffix(name, "-used-percent") {
}
// writeProbeReport writes the result of a probe run.
-func writeProbeReport(out io.Writer, appended int, stop string, records []usageRecord, sizes []int) error {
+func writeProbeReport(out io.Writer, appended int, stop string, records []ModelUsageRecord, sizes []int) error {
var input, cached, output, failed int
for _, record := range records {
input += record.InputTokens
// timeText describes the time per call and how many calls ran at once. A call
// runs from its date for its seconds.
-func timeText(records []usageRecord) string {
+func timeText(records []ModelUsageRecord) string {
type event struct {
at time.Time
delta int
}
// quotaText returns the quota headers of record, sorted by name.
-func quotaText(record usageRecord) string {
+func quotaText(record ModelUsageRecord) string {
names := slices.Sorted(maps.Keys(record.Quota))
parts := make([]string, len(names))
for n, name := range names {
projects / onechat / blobdiff