Agent mode sql nexus pijocoder 051526#527
Closed
nishanthini089 wants to merge 18 commits into
Closed
Conversation
…CPServer_SQLNexus_pijocoder_050126 sh: line 1: w: command not found
…ve/change the top cpu queries
…t script - Consolidated BUILD_SUMMARY.md, GETTING_STARTED.md, HOW_TO_USE.md, and SETUP.md into a single README.md with setup, testing, architecture, MCP protocol flow diagram, and troubleshooting sections - Added --server, --database, and --trusted-connection command-line args to Program.cs so connection details can be passed directly without relying on appsettings.json or environment variables - Fixed stdin EOF handling in ProcessRequests() to break on null ReadLine (pipe close) instead of looping forever, enabling clean exit after piped test input - Updated Test-McpServer.ps1: - Replaced hardcoded paths with script-relative paths - Added interactive prompts for server/database with defaults - Passes --server and --database args to the exe on each call - Implemented Invoke-McpTool helper with Format-Table output - Fixed while loop exit (replaced switch break with $exit flag)
…c, MCP lifecycle) McpTypes.cs - Added NullValueHandling.Ignore to JsonRpcResponse.Result and Error so responses include only result OR error, never both — fixing a strict JSON-RPC 2.0 spec violation that caused Copilot CLI to reject all responses (VS Code was tolerant; CLI was not) Program.cs - Notification handling: requests with null id are now treated as notifications and never produce a response; notifications/initialized is explicitly recognized as a no-op per MCP lifecycle spec - Protocol version negotiation: initialize now echoes back the client's requested protocolVersion instead of hardcoding "2024-11-05" - JSON-RPC batch support: lines starting with '[' are parsed as JArray and each element processed individually - Lazy SQL initialization: DiagnosticAnalyzer is created on first tool call via GetAnalyzer() — server now always completes the MCP handshake even if SQL Server is temporarily unavailable at startup README.md - Updated note to include Copilot CLI alongside VS Code as a supported client - MCP Protocol Flow diagram updated to show notifications/initialized step and protocol version negotiation - Added troubleshooting entry for sqlnexus_MCP showing X in Copilot CLI START_HERE.md - CLI config example updated to include "type":"stdio" (required by CLI)
…d delta query stats
DiagnosticAnalyzer.cs
AnalyzeCpuUsage() — complete rewrite:
- Replaced wait-stats-only implementation with a single Perfmon query that
returns per-sample rows (sample_time, sql_cpu_pct, nonsql_cpu_pct,
system_idle_pct) from CounterData joined by RecordIndex, matching the
Bottleneck Analysis RDL DataSet query exactly
- SET ANSI_NULLS OFF added to match RDL behaviour: prevents the
det.InstanceIndex = @inst_index join from silently returning zero rows
when @inst_index is NULL
- Falls back to tbl_SQL_CPU_HEALTH (ring buffer) when CounterData is absent
- Aggregate summary (max, avg, count above 70%) computed in C# from the
already-fetched rows — no second SQL round-trip
- Consecutive high-CPU run detector: flags any run of 3+ samples above 70%
with run_start, run_end, consecutive_samples, peak_sql_cpu_pct
- sustained_high_cpu_detected boolean gives Copilot an immediate signal
GetTopCpuQueries():
- ReadTrace path: added pct_of_cpu_capacity column — % of total CPU
capacity each query consumed during the collection window, computed as
SUM(CPU) / (cpu_count * collection_duration_ms)
- dm_exec_query_stats path (tbl_Hist_Top10_CPU_Queries_ByQueryHash):
- Delta between first/last snapshot isolates CPU used only during the
collection window (not cumulative since SQL Server startup)
- pct_of_cpu_capacity added using tbl_RUNTIMES for the time window
- OBJECT_ID guard on tbl_RUNTIMES prevents parse-time error when absent
- Added avg_cpu_ms, delta_executions columns; fixed ORDER BY to DESC
- Added ExecuteQueryToDataTable() private helper for methods that compose
multi-section JSON from multiple queries
Program.cs
- Updated analyze_cpu_usage tool description to reflect new Perfmon summary
output: max/avg SQL CPU %, samples above 70%, sustained high CPU detection
…CPServer_SQLNexus_pijocoder_050126
…le catalog DiagnosticAnalyzer.cs: - Add ListNexusTables() method exposing a curated catalog of ~70 analytically significant SQL Nexus tables (dbo + ReadTrace schema) - Each entry carries a plain-English description of the table's source DMV/query, key columns, and how to use it for analysis - At call time, queries sys.tables to annotate each entry with present_in_database so the AI knows which tables were actually captured in this collection - Result includes a discovery_hint pointing to INFORMATION_SCHEMA.TABLES so agents know the catalog is a subset and how to enumerate the rest - Correct tbl_System_Requests description: these are internal system thread rows (session_id < 0) covering checkpoint, lazy writer, log writer and ghost cleanup — not user requests - Correct tbl_SQL_CPU_HEALTH description: confirmed populated from the SQL Server ring buffer health records; used as CPU fallback when Perfmon CounterData is absent - Add tbl_sqlagent_jobs entry (Agent job definitions for incident timeline correlation) - Add tbl_database_options entry (per-database options snapshot) - Add ReadTrace.tblInterestingEvents entry (deadlock graphs, attention events, errors) - Fix tblErrorlog key → tbl_ERRORLOG to match actual table name - Improve tbl_running_drivers description to call out filter drivers (antivirus, backup agents) explicitly Program.cs: - Register list_nexus_tables MCP tool in HandleListTools() - Wire list_nexus_tables case in HandleToolCall() switch - Tool description explicitly states it is a curated subset and embeds the INFORMATION_SCHEMA discovery query for AI agent awareness
…entations - analyze_cpu_usage: remove incorrect reference to tbl_OS_WAIT_STATS / wait type section (SOS_SCHEDULER_YIELD, CXPACKET, CXCONSUMER); describe actual CounterData → tbl_SQL_CPU_HEALTH fallback, per-sample output, and sustained high-CPU run detection (3+ consecutive samples above 70%) - get_top_cpu_queries: replace wrong tbl_NOTABLEACTIVEQUERIES source with accurate dual-path description: ReadTrace.tblBatches (primary) and tbl_Hist_Top10_CPU_Queries_ByQueryHash delta-snapshot (fallback); mention pct_of_cpu_capacity - analyze_io_waits: add LOGBUFFER and ASYNC_IO_COMPLETION to listed wait types; clarify first/last snapshot delta approach - get_collection_time_range: add caveat that it depends on ReadTrace.tblBatches and returns no data for non-ReadTrace collections
- AnalyzeSpinlocks: add is_high_cpu_driver column (1 if spins_per_millisecond_per_CPU > 20000, else 0) - GetCollectionTimeRange: check tbl_RUNTIMES first; fall back to ReadTrace.tblBatches if not present
…tup docs, query scripts
…with MCP tool workflows and skill file references
New tools in DiagnosticAnalyzer.cs + Program.cs: - get_query_execution_details (per-execution drill-down by HashID) - get_wait_type_distribution (request-level wait type frequency) - get_wait_resource_hotspots (hot pages/rows/keys causing contention) - get_wait_heavy_queries (queries with CPU < 80% of duration) - get_statements_in_batch (statement breakdown inside a batch) - get_blocking_chain_tree (recursive CTE blocking hierarchy) - get_lock_summary_by_object (lock contention by table/resource) - get_queries_by_application (queries filtered by ApplicationName) - get_performance_by_application (aggregate perf per ApplicationName) - get_cpu_by_database (CPU breakdown by database) - get_top_queries_by_reads (I/O heavy queries, read pressure) - get_top_queries_by_writes (write heavy / WRITELOG pressure) - get_sql_file_io_stats (per-file avg read/write latency) - get_compilation_stats (SQL Compilations/sec from Perfmon) - get_plan_cache_analysis (cache composition, single-use plans) - get_table_statistics_health (stale stats detection) agent.md tools list updated to include all 34 MCP tools.
…dd read+search aliases for skill file access
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
3 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Agent created on top of the MCP Server branch with additional SQL Nexus queries as tools + diagnostic skills from SQL Nexus wiki added