fix(llm): honour configured max_tokens on AI analysis paths

[dpage]

Fixes #399

Problem

Three AI-analysis paths hardcoded a small output-token cap that ignored the operator's configured llm.max_tokens:

Path	Old value
Estate summary (overview/generator.go)	llmMaxTokens = 512
Server-info AI analysis (api/server_info_handlers.go)	llmAnalysisMaxTokens = 512
Alerter reasoning (alerter/.../llm/reasoning.go)	pgllm.Int(500)

Reasoning models served through an OpenAI-compatible endpoint charge their thinking tokens against the same max_tokens budget as the answer. A 512-token budget is routinely consumed in full by the thinking block, so the model never emits a text block.

extractTextFromResponse collected only pgllm.BlockText blocks and returned an empty string, which callers treated as a successful generation. The estate summary therefore rendered blank with no error. This reproduces against a local llama.cpp server and was originally hit while building an all-in-one Docker stack.

The constants predate the pgedge-go-llm-lib migration and the BuildClientOptions extraction; both of those were behaviour-preserving refactors that carried the values through unchanged.

Changes

• Resolve the output budget from configuration on all three paths, falling back to 4096 when the setting is unset, zero, or negative.
• BuildClientOptions now resolves max-tokens internally via the new nil-safe AnalysisMaxTokens() rather than accepting it as a parameter. A max-tokens parameter is exactly what let the two analysis call sites drift into hardcoding their own constants; removing it makes divergence impossible. Temperature stays a parameter because the two paths legitimately tune it.
• Treat a response carrying no usable text as an error (ErrNoTextContent) rather than an empty summary, so the condition surfaces instead of silently caching a blank panel. The error message names the likely cause and the setting to change.
• Map that condition to 502 on GET /api/v1/server-info/{id}/ai-analysis, documented in openapi.go and the regenerated openapi.json.
• Add llm.max_tokens to the alerter, which had no equivalent setting, with a 4096 default plus the config merge and defaulting passes.

Note on the exported API

BuildClientOptions changes signature from (maxTokens int, temperature float64) to (temperature float64). Both call sites are internal to this repository and are updated here.

Behaviour deliberately left unchanged

The transient failure paths in getAIAnalysis (nil LLM config, no databases, client construction failure, chat error) still return 200 with a null body meaning "analysis unavailable". Only the "model answered but produced nothing usable" case becomes an error. Turning a provider outage into a 502 would reach well beyond this issue.

The web client's existing .catch in ServerInfoDialog.tsx already logs and leaves the panel empty, so no client change is needed.

Testing

Table-driven tests cover each path for: configured value used; fallback on unset, zero, and negative; a small configured value still honoured; empty, whitespace-only, reasoning-block-only, and nil responses all yielding an error; non-empty responses passing through unchanged; and the max_tokens value actually observed on the wire, decoded from a stub server's request body.

Per-function line coverage on every touched function (verified with go tool cover -func):

llmproxy: AnalysisMaxTokens 100.0%  BuildClientOptions 100.0%
overview: extractTextFromResponse 100.0%  generateSummaryFromPrompt 100.0%  createLLMClient 100.0%
api:      getAIAnalysis 100.0%  writeAIAnalysisResponse 100.0%  createLLMClient 100.0%
          parseDatabaseAnalysisResponse 94.1%  handleServerInfoAI 91.3%
alerter:  Classify 100.0%  newLibReasoning 100.0%  NewReasoningProvider 100.0%
          SetLLMDefaults 100.0%  NewConfig 100.0%  LoadFromFile 100.0%

The single uncovered statement in parseDatabaseAnalysisResponse is in the pre-existing numbered-list-prefix stripper, untouched here.

gofmt -l is clean. cd server && make test and cd alerter && make test both pass. go test ./internal/api/ -run OpenAPI -v passes.

make test-all could not complete locally: the collector's lint step fails because the installed golangci-lint is built with Go 1.25 while the config targets Go 1.26.2. This reproduces on a pristine tree with all changes stashed, so it is pre-existing and unrelated to this change. go vet ./... is clean on both modules as a substitute; CI is the real gate.

🤖 Generated with Claude Code

https://claude.ai/code/session_01Piu2z26Jjg1fuPrkSJ5aFq

---

Generated by Claude Code

[coderabbitai]

Warning

Review limit reached

You’ve reached a temporary PR review limit under our Fair Usage Limits Policy.

Your recent review volume is higher than typical usage, so adaptive limits are currently applied.

Next review available in: 15 minutes

Enable usage-based reviews in Billing to review now. Otherwise, wait until the next included review is available.
You're only billed for reviews past your plan's rate limits ($0.25/file).

How can I continue?

After more reviews become available, a review can be triggered using the @coderabbitai review command as a PR comment. Alternatively, push new commits to this PR.

To avoid repeated limits, reduce automatic review volume by pausing incremental auto-reviews earlier, using label-based review opt-in, excluding WIP or generated PR titles, or requesting reviews manually when the PR is ready. If your team needs uninterrupted high-volume reviews, an organization admin can enable usage-based reviews.

How do review limits work?

CodeRabbit enforces per-developer PR review limits for each organization. Most developers receive the normal plan review availability.

For paid Pro and Pro+ PR reviews, CodeRabbit uses adaptive limits for sustained high-volume activity. When a developer's recent PR review activity reaches the 95th percentile or higher among CodeRabbit users, additional reviews become available more gradually as earlier reviews age out of the rolling window.

Please refer docs for additional details.

Review details

⚙️ Run configuration

Configuration used: Path: .coderabbit.yaml

Review profile: CHILL

Plan: Pro

Run ID: 6d21975d-9273-4b7e-9836-20f4dc22e425

📥 Commits

Reviewing files that changed from the base of the PR and between 19c645d and f75d571.

📒 Files selected for processing (20)

• .claude/golang-expert/testing-strategy.md
• alerter/src/internal/config/config.go
• alerter/src/internal/config/config_test.go
• alerter/src/internal/llm/llm.go
• alerter/src/internal/llm/reasoning.go
• alerter/src/internal/llm/reasoning_test.go
• docs/admin-guide/api/openapi.json
• docs/admin-guide/api/server-info.md
• docs/changelog.md
• docs/getting-started/configuration/alerter.md
• docs/getting-started/configuration/server.md
• examples/ai-dba-alerter.yaml
• examples/ai-dba-server.yaml
• server/src/internal/api/openapi.go
• server/src/internal/api/server_info_handlers.go
• server/src/internal/api/server_info_handlers_test.go
• server/src/internal/llmproxy/proxy.go
• server/src/internal/llmproxy/proxy_test.go
• server/src/internal/overview/generator.go
• server/src/internal/overview/generator_test.go

---

_{Comment @coderabbitai help to get the list of available commands.}

[codacy-production]

Up to standards ✅

🟢 Issues 0 issues

Results:
0 new issues

View in Codacy

🟢 Metrics 50 complexity · 2 duplication

Metric	Results
Complexity	50
Duplication	2

View in Codacy

_{NEW Get contextual insights on your PRs based on Codacy's metrics, along with PR and Jira context, without leaving GitHub. Enable AI reviewer
TIP This summary will be updated as you push new changes.}