Tune tracker workers and bound agent budgets

The tracker-mapping and common-pattern-enrichment workers ran with the
kit/worker defaults (interval 10s, max-concurrency 5 each) and dropped
the resolved per-agent max-tokens/temperature, so up to ten LLM
pipelines could run unbounded on one OpenAI client. The mapping worker
also held a FOR UPDATE transaction across the LLM and Firecrawl calls
while its DB search tools acquired a second pooled connection, risking
pool exhaustion under concurrency.

Plumb max-tokens, temperature, agent timeout, and per-worker max-turns
through TrackerAgentsConfig and DisambiguationConfig into all three
agent builders, replacing the hard-coded constants with config-fed
fields and package fallbacks. Expose worker interval, concurrency,
stale-after, agent timeout, and max-turns as config (env, Helm values,
deployment template) mirroring the evidence-describer pattern, and
apply them at registration.

Refactor Process into deterministic-read, agent (no transaction), and
persist phases so neither the mapping agent nor disambiguation runs
inside an open transaction, removing the row locks held across network
latency and the nested-connection pressure.

Signed-off-by: Émile Ré <emile@probo.com>
This commit is contained in:
Émile Ré
2026-05-29 16:37:32 +02:00
parent 50c5454681
commit 55302d18f0
17 changed files with 818 additions and 339 deletions

View File

@@ -41,13 +41,17 @@ type (
LLMAgentConfig = probodconfig.LLMAgentConfig
EvidenceDescriberConfig = probodconfig.EvidenceDescriberConfig
AgentsConfig = probodconfig.AgentsConfig
MailerConfig = probodconfig.MailerConfig
SMTPConfig = probodconfig.SMTPConfig
NotificationsConfig = probodconfig.NotificationsConfig
WebhookConfig = probodconfig.WebhookConfig
OIDCProviderConfig = probodconfig.OIDCProviderConfig
PgConfig = probodconfig.PgConfig
SAMLConfig = probodconfig.SAMLConfig
SCIMBridgeConfig = probodconfig.SCIMBridgeConfig
SlackConfig = probodconfig.SlackConfig
TrackerMappingWorkerConfig = probodconfig.TrackerMappingWorkerConfig
CommonPatternEnrichmentWorkerConfig = probodconfig.CommonPatternEnrichmentWorkerConfig
MailerConfig = probodconfig.MailerConfig
SMTPConfig = probodconfig.SMTPConfig
NotificationsConfig = probodconfig.NotificationsConfig
WebhookConfig = probodconfig.WebhookConfig
OIDCProviderConfig = probodconfig.OIDCProviderConfig
PgConfig = probodconfig.PgConfig
SAMLConfig = probodconfig.SAMLConfig
SCIMBridgeConfig = probodconfig.SCIMBridgeConfig
SlackConfig = probodconfig.SlackConfig
)

View File

@@ -725,7 +725,14 @@ func (impl *Implm) Run(
},
)
trackerMappingWorker := cookiebanner.NewTrackerMappingWorker(pgClient, l, trackerAgentsCfg, thirdPartyDisambiguationCfg)
trackerMappingWorker := cookiebanner.NewTrackerMappingWorker(
pgClient,
l,
trackerAgentsCfg,
thirdPartyDisambiguationCfg,
worker.WithInterval(time.Duration(impl.cfg.TrackerMappingWorker.Interval)*time.Second),
worker.WithMaxConcurrency(impl.cfg.TrackerMappingWorker.MaxConcurrency),
)
trackerMappingWorkerCtx, stopTrackerMappingWorker := context.WithCancel(context.Background())
wg.Go(
@@ -742,7 +749,17 @@ func (impl *Implm) Run(
stopCommonPatternEnrichmentWorker := func() {}
if trackerAgentsCfg.LLMClient != nil {
commonPatternEnrichmentWorker := cookiebanner.NewCommonPatternEnrichmentWorker(pgClient, l, trackerAgentsCfg)
enrichmentCfg := trackerAgentsCfg
enrichmentCfg.AgentTimeout = time.Duration(impl.cfg.CommonPatternEnrichmentWorker.AgentTimeout) * time.Second
commonPatternEnrichmentWorker := cookiebanner.NewCommonPatternEnrichmentWorker(
pgClient,
l,
enrichmentCfg,
time.Duration(impl.cfg.CommonPatternEnrichmentWorker.StaleAfter)*time.Second,
worker.WithInterval(time.Duration(impl.cfg.CommonPatternEnrichmentWorker.Interval)*time.Second),
worker.WithMaxConcurrency(impl.cfg.CommonPatternEnrichmentWorker.MaxConcurrency),
)
var commonPatternEnrichmentWorkerCtx context.Context

View File

@@ -16,6 +16,7 @@ package probod
import (
"fmt"
"time"
"github.com/prometheus/client_golang/prometheus"
"go.gearno.de/kit/log"
@@ -57,15 +58,34 @@ func (impl *Implm) buildTrackerAgentsConfig(
return cookiebanner.TrackerAgentsConfig{}, thirdparty.DisambiguationConfig{}, fmt.Errorf("cannot resolve tracker mapping agent client: %w", err)
}
mappingWorkerCfg := impl.cfg.TrackerMappingWorker
enrichmentWorkerCfg := impl.cfg.CommonPatternEnrichmentWorker
// The mapping and enrichment agents share one config slot but run
// from separate workers with separate max-turns. AgentTimeout here
// carries the mapping worker's value (also reused by the
// disambiguation agent); the enrichment worker overrides it on its
// own copy at registration.
trackerAgentsCfg := cookiebanner.TrackerAgentsConfig{
LLMClient: llmClient,
Model: agentCfg.ModelName,
FirecrawlAPIKey: impl.cfg.Agents.Tools.FirecrawlAPIKey,
LLMClient: llmClient,
Model: agentCfg.ModelName,
FirecrawlAPIKey: impl.cfg.Agents.Tools.FirecrawlAPIKey,
MaxTokens: agentCfg.MaxTokens,
Temperature: agentCfg.Temperature,
AgentTimeout: time.Duration(mappingWorkerCfg.AgentTimeout) * time.Second,
MappingMaxTurns: mappingWorkerCfg.AgentMaxTurns,
EnrichmentMaxTurns: enrichmentWorkerCfg.AgentMaxTurns,
}
// The disambiguation agent emits a single id plus a short rationale,
// so it keeps its own smaller token budget (left unset here) rather
// than inheriting the mapping agent's. It shares the mapping worker's
// timeout.
disambiguationCfg := thirdparty.DisambiguationConfig{
LLMClient: llmClient,
Model: agentCfg.ModelName,
LLMClient: llmClient,
Model: agentCfg.ModelName,
Temperature: agentCfg.Temperature,
Timeout: time.Duration(mappingWorkerCfg.AgentTimeout) * time.Second,
}
return trackerAgentsCfg, disambiguationCfg, nil