diff --git a/docs/admin/config/settings.mdx b/docs/admin/config/settings.mdx index 34010fabf..d3085185b 100644 --- a/docs/admin/config/settings.mdx +++ b/docs/admin/config/settings.mdx @@ -98,7 +98,6 @@ Settings options and their default values are shown below. "batchChangeAgentsDisabledDefaultSkills": [], "batchChangeAgentsGuidance": [], "boostRelevantRepositories": true, - "changesetsReviewPage": false, "clientSearchResultRanking": "by-zoekt-ranking", "codeInsightsRepoUI": "single-search-query", "deepSearchGuidance": [], diff --git a/docs/admin/config/site-config.mdx b/docs/admin/config/site-config.mdx index ed881e7b6..2296cfb65 100644 --- a/docs/admin/config/site-config.mdx +++ b/docs/admin/config/site-config.mdx @@ -195,11 +195,6 @@ All advanced configuration options and their default values are shown below. "commands": false }, - // Enables the computation of contributor statistics per author and repository. Will all commits of each repository initially, and then work on deltas. - // Other example values: - // - true - "contributorsDataEnabled": true, - // Required when using any of the native code host integrations for Phabricator, GitLab, or Bitbucket Server. It is a space-separated list of allowed origins for cross-origin HTTP requests which should be the base URL for your Phabricator, GitLab, or Bitbucket Server instance. // Other example values: // - "https://my-phabricator.example.com https://my-bitbucket.example.com https://my-gitlab.example.com" diff --git a/docs/ai/models.mdx b/docs/ai/models.mdx index 09315fa5f..bfc7ae7ec 100644 --- a/docs/ai/models.mdx +++ b/docs/ai/models.mdx @@ -9,7 +9,7 @@ This reference lists the models currently used by [Sourcegraph AI features](/ai) | --------------------- | -------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- | | Claude Haiku 5.5 | Deep Search titles, [Code Finder](/api/mcp#code-finder), [Smart Hover](/code-navigation/smart-hover), Agentic Batch Changes titles, Agentic Batch Changes workspace analysis, Agentic Batch Changes schema help, Coding agent low Finder, Coding agent medium Finder, Coding agent high Finder | | Claude Opus 5.5 | [Agentic Batch Changes](/agentic-batch-changes), Agentic Batch Changes compaction | -| Claude Sonnet 5 | [Deep Search](/deep-search) | +| Claude Sonnet 5.5 | [Deep Search](/deep-search) | | Gemini 3.5 Flash-Lite | [Query Assist](/code-search/features#query-assist), Agentic Batch Changes repository search | | Gemini 4 Argon | Coding agent high | | GPT Transcribe | [Dictation](https://sourcegraph.com/changelog/agent-dictation) | diff --git a/docs/self-hosted/observability/alerts.mdx b/docs/self-hosted/observability/alerts.mdx index f54bebb6a..8a272900c 100644 --- a/docs/self-hosted/observability/alerts.mdx +++ b/docs/self-hosted/observability/alerts.mdx @@ -7544,20 +7544,21 @@ Custom query for critical alert: `max(((sum by (workload, agent) (increase(src_a **Descriptions** -- critical deepsearch: 5+ truncated questions with no answer produced over 15m for 10m0s +- warning deepsearch: 5+ truncated questions with no answer produced over 15m for 10m0s **Next steps** -- Search worker logs for: "answer truncated due to output token limit with no response generated" to find affected questionIDs. - The output token budget was consumed before any answer text was produced, usually by reasoning tokens, so affected users got nothing back. -- Check if a recent model change or reasoning effort change caused increased thinking token usage. +- Check whether a Deep Search model, reasoning effort, or output token limit change, or a deploy, landed shortly before the alert started. If one did, it is the likely cause: roll it back. +- Otherwise, search worker logs for: "answer truncated due to output token limit with no response generated" and group by the logged userID. Bursts so far have come from one user or API client sending a batch of questions, cleared on their own within about 20 minutes, and needed no action. +- If affected users are spread out, check for an upstream model problem: compare per-model stream duration (`sum by (model) (rate(src_completions_stream_duration_seconds_sum[10m])) / sum by (model) (rate(src_completions_stream_duration_seconds_count[10m]))`) with the hours before, and check the model provider status page. - Compare against `deepsearch_response_truncated`: elevated truncations alone are usually users requesting very large outputs and are not a regression. - More help interpreting this metric is available in the [dashboards reference](/self-hosted/observability/dashboards#deepsearch-deepsearch_response_truncated_no_answer). - **Silence this alert:** If you are aware of this alert and want to silence notifications for it, add the following to your site configuration and set a reminder to re-evaluate the alert: ```json "observability.silenceAlerts": [ - "critical_deepsearch_deepsearch_response_truncated_no_answer" + "warning_deepsearch_deepsearch_response_truncated_no_answer" ] ``` @@ -7566,7 +7567,7 @@ Custom query for critical alert: `max(((sum by (workload, agent) (increase(src_a
Technical details -Custom query for critical alert: `max(((sum(increase(src_deepsearch_question_outcome_total{outcome="response_truncated_no_answer"\}[15m]))) and (sum(increase(src_deepsearch_question_outcome_total\{outcome="response_truncated_no_answer"\}[5m])) >= 2) and (sum(increase(src_deepsearch_question_outcome_total\{outcome="response_truncated_no_answer"}[15m])) / sum(increase(src_deepsearch_question_outcome_total[15m])) >= 0.05) and (sum(increase(src_deepsearch_question_outcome_total[15m])) >= 20)) >= 5)` +Custom query for warning alert: `max(((sum(increase(src_deepsearch_question_outcome_total{outcome="response_truncated_no_answer"\}[15m]))) and (sum(increase(src_deepsearch_question_outcome_total\{outcome="response_truncated_no_answer"\}[5m])) >= 2) and (sum(increase(src_deepsearch_question_outcome_total\{outcome="response_truncated_no_answer"}[15m])) / sum(increase(src_deepsearch_question_outcome_total[15m])) >= 0.05) and (sum(increase(src_deepsearch_question_outcome_total[15m])) >= 20)) >= 5)`
diff --git a/docs/self-hosted/observability/dashboards.mdx b/docs/self-hosted/observability/dashboards.mdx index 64f43bef8..0bd21da90 100644 --- a/docs/self-hosted/observability/dashboards.mdx +++ b/docs/self-hosted/observability/dashboards.mdx @@ -38051,7 +38051,7 @@ sum by (outcome) (increase(src_deepsearch_question_outcome_total{billable="true"

Truncated questions with no answer produced over 15m

-Number of deep search questions that hit the output token limit before producing any answer text in the last 15 minutes. Affected users received nothing, so this is the class that indicates a model or prompt regression rather than an oversized user request. +Number of deep search questions that hit the output token limit before producing any answer text in the last 15 minutes. Affected users received nothing. A rise that persists after a model, reasoning effort, or prompt change points to a regression; short bursts from one user sending a batch of questions have so far needed no action. Refer to the [alerts reference](/self-hosted/observability/alerts#deepsearch-deepsearch_response_truncated_no_answer) for 1 alert related to this panel.