Repository navigation
Releases
Stackish 0.1.0 Preview 1 — Windows and Mac
Mac downloads withdrawn: The published Mac bundles had invalid signatures and could show “Stackish is damaged.” They were not verified through the downloaded-app launch path. A packaging fix is in progress; no replacement Mac download is being claimed ready. Developer ID signing and notarization are still required for normal Gatekeeper acceptance. The Windows download is unchanged.
Stackish's first downloadable preview: one Windows/Mac app that generates a mixed SQL Server workload and compares before/after tuning throughput.
Downloads
- Windows x64:
Stackish-win-x64.zip. Extract the entire ZIP and openStackish.exe; keep its companion files together. - Mac with Apple Silicon: withdrawn pending a verified distribution build.
- Mac with an Intel processor: withdrawn pending a verified distribution build.
SHA256SUMS.txtcontains the SHA-256 checksum for the remaining Windows archive.
These builds include .NET; users do not need an SDK. They are unsigned previews, not signed/notarized installers. Windows or macOS may warn or block them under local security policy. Linux is not included. Native Windows and Intel Mac GUI validation remains outstanding.
Set up the database first
Use a disposable 2024-04 Stack Overflow restore on SQL Server 2025 or later, compatibility level 170 or later. From the repository root, deploy with sqlcmd -S YOUR_SERVER -U YOUR_LOGIN -d YOUR_DATABASE -b -i sql/deploy.sql. Keep the password out of the command and repository. Deployment installs the procedures/function and creates missing supporting indexes; it does not restore or reset the database.
Open the app, enter the database connection, click Test connection, choose concurrency, then Start workload. The workload commits writes. Click Stop and wait for Stopped before restoring or cutting over. The app is SQL Server-only; porting it to another database is part of the migration exercise.
See the setup and operating instructions. The desktop/core code is identical to reviewed main commit 2afd3ce; this release includes the migration-focused README from a4a3f80.
Validation
All 12 core tests passed. The archives passed integrity, expected runtime-file, executable-format, and Mac executable-permission checks. Existing platform CI passed Windows x64, Mac arm64, and Mac x64 builds and offline previews. The packaged Apple Silicon app also opened natively on macOS, exposed the expected accessible control names, and handled incomplete connection settings without crashing.
Continuous 2024-04 testbed run
The actual desktop workload engine ran for 30 minutes with eight concurrent operations, across all twelve default operation groups. This was a completed run with SQL errors, not a clean pass:
- 407 completed reads and 339 applied writes during measurement (746 calls total; 0.414 calls/sec overall).
- Zero Conflict, NoChange, or NotFound outcomes; no write failures were observed.
- Three ten-minute search timeouts and one search deadlock victim. The runner recorded the errors and continued. The evidence client correctly returned a nonzero exit status because errors occurred.
- The longest stretch without a completed call was 185 seconds. Expensive searches occupied the workers and created gaps in new writes. A server sample showed query-memory waits and parallel-query waits, with no blocking session reported.
- Stop completed in 30.44 seconds: one read and one applied vote finished during drain, and six remaining reads were canceled. Drain completions were excluded from measured throughput. No Stackish sessions, active requests, or open transactions remained afterward.
- Successful call durations (including the two drain completions): median 113 ms, 95th percentile 28.9 seconds, maximum 597 seconds. These are observed client durations on this testbed, not a performance promise.
Workload writes were committed as intended; the database was not restored, dropped, or reset. SQLQueryStress itself was not run. The native Windows/Intel Mac GUI and longer-duration/unattended operation remain unvalidated. Do not infer that continuous operation means a steady rate of successful writes.