SQL Server Performance Triage

Works against any SQL Server instance (2005 → current). The bundled script scripts/Invoke-SqlPerfTriage.ps1 collects a read-only diagnostic snapshot and emits JSON; the agent (you) writes the analysis. The script never modifies server or database state.

SCRIPT = this skill's scripts/Invoke-SqlPerfTriage.ps1. Interpretation guide in REFERENCE.md. This script is vendored identically in the NAV 2009 and AX 2012 sibling skills — if the target database belongs to one of those applications, prefer the sibling skill for its application-specific interpretation.

Permissions & auth

• Default is Windows integrated auth. Pass -SqlCredential (Get-Credential) for SQL auth — never put a password on the command line.
• Needs VIEW SERVER STATE on the instance and VIEW DATABASE STATE (or db_owner) in the target database. The trace-flag sub-check may need sysadmin and degrades gracefully.
• Connections are unencrypted by default; add -Encrypt if the instance has a valid certificate.

How to run

Always run with pwsh. Parse the JSON it prints on stdout.

Want	Pass
Full snapshot of a DB	-ServerInstance SQLSRV01 -Database 'AppDB'
Named instance	-ServerInstance 'SQLSRV01\INST01'
Server-level only (no DB)	omit -Database (DB-scoped sections report skipped)
Only some sections	-Sections waits,blocking,deadlocks
SQL auth	-SqlCredential (Get-Credential)
Slow big DB	-QueryTimeout 300 (fragmentation scan is the slow one; or drop it from -Sections)
Save full report	-OutFile C:\path\triage.json

Sections: server, database, waits, top_queries, missing_indexes, unused_indexes, blocking, deadlocks, sift, fragmentation, stats, largest_tables (default all). The sift section is NAV-specific and harmlessly reports 0 views on non-NAV databases.

Examples:

# Full triage of one database
pwsh -File SCRIPT -ServerInstance SQLSRV01 -Database 'AppDB' -OutFile C:\ops\triage.json

# "The server is slow right now" — live picture only
pwsh -File SCRIPT -ServerInstance SQLSRV01 -Sections waits,blocking,deadlocks,top_queries

Output contract

• Without -OutFile → full JSON on stdout.
• With -OutFile → full JSON to the file; a compact summary (per-section status + row counts) on stdout. Prefer -OutFile for full snapshots; read back only what you need.

Top level: status (ok/partial/error) and sections, each { status: ok|error|skipped, data|error|reason }. Timestamps in data are SQL Server local time; deadlock timestamps are UTC. Full schema in REFERENCE.md.

What you (the agent) do with the result

1. Run the script, parse the JSON.
2. Triage in this order: live blocking → deadlocks → waits (what is the server actually waiting on?) → top queries (who causes it?) → indexes/stats/config (why?). Use the interpretation table in REFERENCE.md.
3. Lead with the 2–4 findings that matter, each as: evidence (numbers) → likely cause → concrete next action. Don't recite every section.
4. Fail loud on coverage: name any error/skipped section. Wait stats and index usage are cumulative since instance restart — state the window (server.system.sqlserver_start_time) before drawing conclusions.
5. Before recommending index changes, ask whether an application owns the schema (ERP systems like Dynamics NAV/AX drop out-of-band indexes on synchronization) — if yes, switch to the application-specific sibling skill.

Errors

• Cannot connect → wrong instance name, SQL Browser off for named instances, firewall, auth. Check with Test-NetConnection <server> -Port 1433.
• Section error: VIEW SERVER STATE permission was denied → request the grant; other sections still ran.
• deadlocks unsupported on SQL Server 2005 (no Extended Events) — offer trace flag 1222 as the manual alternative.