@mndrk/agx 2.0.38 → 2.0.40

package/cloud-runtime/standalone/docs/baseline/critical-queries.md

@@ -0,0 +1,105 @@
+# Performance Baseline: PostgreSQL Query Patterns
+
+**Captured:** 2026-02-21
+**Database:** PostgreSQL (Docker, `agx-postgres`)
+**Connection:** `postgresql://agx:agx@localhost:55432/agx`
+**Dataset:** ~84 tasks, ~2253 task_logs, ~4144 graph_nodes, ~4245 graph_edges
+
+> **Note:** `pg_stat_statements` is installed but not loaded in `shared_preload_libraries`,
+> so per-query latency histograms (p50/p99) are unavailable. Baselines below use
+> `EXPLAIN ANALYZE` single-run execution times and `pg_stat_user_tables`/`pg_stat_user_indexes`
+> cumulative counters.
+
+---
+
+## Critical Queries by Index Scan Frequency
+
+| # | Query Pattern | Index Used | idx_scan | Exec Time (ms) | Plan |
+|---|--------------|------------|----------|-----------------|------|
+| 1 | `SELECT * FROM tasks WHERE id = $1` | `tasks_pkey` | 72,298 | **0.084** | Index Scan |
+| 2 | `SELECT * FROM tasks WHERE status='queued' ORDER BY priority DESC, created_at LIMIT 1` | `idx_tasks_queue` | 27,975 | **0.426** | Index Scan + Incremental Sort |
+| 3 | `SELECT * FROM task_logs WHERE task_id = $1 ORDER BY created_at` | `idx_task_logs_task` | 15,291 | **1.330** | Bitmap Heap Scan (180 rows) |
+| 4 | `SELECT * FROM tasks WHERE status = $1 AND retry_count < $2` | `idx_tasks_status_retry` | 9,196 | **0.038** | Index Scan |
+| 5 | `SELECT * FROM tasks WHERE project_id = $1` | `idx_tasks_project` | 3,504 | **0.196** | Seq Scan (small table) |
+| 6 | `SELECT * FROM graph_edges WHERE graph_id = $1` | `idx_graph_edges_graph_id` | 3,331 | **0.314** | Bitmap Heap Scan (56 rows) |
+| 7 | `SELECT * FROM learnings WHERE user_id = $1` | `idx_learnings_user` | 2,652 | N/A | Index Scan |
+| 8 | `SELECT * FROM graph_nodes WHERE graph_id = $1` | `idx_graph_nodes_graph_id` | 2,300 | **0.227** | Bitmap Heap Scan (57 rows) |
+| 9 | `SELECT * FROM project_repos WHERE project_id = $1` | `idx_project_repos_project` | 2,290 | <0.1 | Index Scan |
+| 10 | `SELECT * FROM task_comments WHERE task_id = $1 AND deleted_at IS NULL ORDER BY created_at` | `idx_task_comments_not_deleted` | 1,451 | **0.200** | Bitmap Heap Scan (10 rows) |
+| 11 | `SELECT * FROM learnings WHERE scope = $1` | `idx_learnings_scope` | 1,238 | <0.1 | Index Scan |
+| 12 | `SELECT * FROM user_settings WHERE user_id = $1` | `user_settings_pkey` | 1,176 | **0.066** | Index Scan |
+| 13 | `SELECT * FROM workflow_nodes WHERE name = $1` | `idx_workflow_nodes_name` | 1,110 | <0.1 | Index Scan |
+| 14 | `SELECT * FROM graph_nodes WHERE graph_id = $1 AND node_id = $2` | `graph_nodes_pkey` | 1,097 | **0.080** | Index Scan |
+| 15 | `SELECT * FROM user_secrets WHERE user_id = $1` | `user_secrets_pkey` | 656 | <0.1 | Index Scan |
+
+---
+
+## Seq Scan Hot Spots
+
+| Table | seq_scan | idx_scan | Concern |
+|-------|----------|----------|---------|
+| `tasks` | 16,757 | 113,333 | High seq_scan count despite indexes — likely `SELECT *` without indexed filter |
+| `execution_graphs` | 14,125 | 146 | **Missing index on `task_id`** — 14k seq scans vs 146 idx scans |
+| `projects` | 4,502 | 0 | No index usage — small table, seq scan acceptable |
+| `graph_events` | 2,532 | 191 | Mostly seq scanned |
+| `project_repos` | 2,002 | 2,290 | Balanced |
+
+**Action item:** `execution_graphs.task_id` needs an index — 14k seq scans is the biggest optimization opportunity.
+
+---
+
+## Write-Heavy Tables
+
+| Table | Inserts | Updates | Deletes | Total Writes | Approx QPS* |
+|-------|---------|---------|---------|--------------|-------------|
+| `graph_edges` | 5,747 | 0 | 5,469 | **11,216** | ~0.13 |
+| `graph_nodes` | 4,076 | 36 | 3,886 | **7,998** | ~0.09 |
+| `task_logs` | 2,445 | 0 | 193 | **2,638** | ~0.03 |
+| `tasks` | 146 | 1,714 | 120 | **1,980** | ~0.02 |
+| `execution_graphs` | 146 | 545 | 120 | **811** | ~0.01 |
+| `task_audit_log` | 232 | 146 | 0 | **378** | <0.01 |
+| `task_run_history` | 124 | 0 | 30 | **154** | <0.01 |
+| `task_comments` | 91 | 0 | 36 | **127** | <0.01 |
+
+*\*QPS estimated over ~24h uptime; this is a dev/local instance, not production-scale traffic.*
+
+**Write patterns:**
+- `graph_edges` and `graph_nodes` are insert-heavy with frequent deletes (graph rebuilds)
+- `tasks` is update-heavy (status transitions, 1,714 updates vs 146 inserts)
+- `task_logs` is append-only (2,445 inserts, minimal deletes)
+
+---
+
+## Table Sizes
+
+| Table | Total Size | Live Rows | Dead Rows |
+|-------|-----------|-----------|-----------|
+| `task_logs` | 110 MB | 2,253 | 194 |
+| `graph_migration_backups` | 13 MB | 0 | 0 |
+| `graph_nodes` | 11 MB | 4,144 | 204 |
+| `tasks` | 7.4 MB | 84 | 17 |
+| `task_comments` | 5.2 MB | 55 | 36 |
+| `graph_edges` | 1.8 MB | 4,245 | 610 |
+| `task_audit_log` | 1.2 MB | 1,807 | 196 |
+
+Total DB footprint: ~150 MB (dominated by `task_logs` at 110 MB).
+
+---
+
+## Lock & Retry Behavior
+
+- **Optimistic locking** via `version` column on `tasks` table (trigger: `agx.increment_task_version()`)
+- **Rate limiting** via `agx.check_rate_limit()` PL/pgSQL function
+- **No advisory locks** or explicit row-level locking observed in the query builder
+- **Retry pattern**: `idx_tasks_status_retry` (status='failed' + retry_count < N) — 9,196 scans
+
+---
+
+## Key Observations for SQLite Migration
+
+1. **All queries are sub-millisecond to ~1.3ms** — SQLite should match or beat this at current dataset sizes
+2. **graph_edges/graph_nodes churn** is the highest-write pattern — test SQLite WAL mode under concurrent graph rebuilds
+3. **execution_graphs needs a task_id index** even in current Postgres setup
+4. **JSON columns** (26 tracked) — verify SQLite JSON functions cover the `@>` containment operator used in GIN-indexed queries
+5. **UPSERT with ON CONFLICT** — verify SQLite supports the same conflict resolution semantics
+6. **Total dataset is small** (~150 MB) — well within SQLite's comfort zone

package/cloud-runtime/standalone/docs/baseline/lock-metrics.json

@@ -0,0 +1,21 @@
+{
+  "benchmark": "lock_contention_baseline",
+  "database": "agx",
+  "timestamp": "2026-02-21T22:15:30Z",
+  "sample_period_seconds": 10,
+  "deadlock_count_during_sample": 0,
+  "deadlock_count_cumulative": 0,
+  "conflicts_during_sample": 0,
+  "stats_reset": "never",
+  "lock_waits_avg_concurrent": 0,
+  "lock_waits_max_concurrent": 0,
+  "lock_summary_snapshot": [{"mode" : "AccessShareLock", "count" : 1, "granted" : true}],
+  "config": {
+    "deadlock_timeout_ms": 1000,
+    "lock_timeout_ms": 0,
+    "log_lock_waits": "off"
+  },
+  "retry_behavior_notes": "No application-level retry metrics instrumented. Postgres serialization failures (40001) would surface as app errors. Lock contention is minimal in current workload.",
+  "pg_host": "localhost",
+  "pg_port": "55432"
+}

package/cloud-runtime/standalone/docs/baseline/read-latency.json

@@ -0,0 +1,146 @@
+{
+  "benchmark": "read-latency",
+  "captured_at": "2026-02-21T22:29:16Z",
+  "database": "postgresql",
+  "connection": "postgresql://agx:agx@localhost:55432/agx",
+  "iterations_per_query": 50,
+  "method": "EXPLAIN (ANALYZE, FORMAT JSON) repeated sampling",
+  "note": "pg_stat_statements not loaded; using repeated EXPLAIN ANALYZE for percentile estimation",
+  "queries": [
+    {
+      "query_id": 1,
+      "name": "tasks_by_id",
+      "samples": 50,
+      "p50_ms": 0.0970,
+      "p90_ms": 0.1640,
+      "p99_ms": 0.5850,
+      "avg_ms": 0.1169
+    },
+    {
+      "query_id": 2,
+      "name": "next_queued_task",
+      "samples": 50,
+      "p50_ms": 0.1870,
+      "p90_ms": 0.2770,
+      "p99_ms": 0.3290,
+      "avg_ms": 0.1701
+    },
+    {
+      "query_id": 3,
+      "name": "task_logs_by_task_id",
+      "samples": 50,
+      "p50_ms": 0.5260,
+      "p90_ms": 0.7720,
+      "p99_ms": 1.2380,
+      "avg_ms": 0.5173
+    },
+    {
+      "query_id": 4,
+      "name": "failed_tasks_for_retry",
+      "samples": 50,
+      "p50_ms": 0.0870,
+      "p90_ms": 0.1160,
+      "p99_ms": 0.1500,
+      "avg_ms": 0.0836
+    },
+    {
+      "query_id": 5,
+      "name": "tasks_by_project",
+      "samples": 50,
+      "p50_ms": 0.1820,
+      "p90_ms": 0.2110,
+      "p99_ms": 0.2630,
+      "avg_ms": 0.1677
+    },
+    {
+      "query_id": 6,
+      "name": "graph_edges_by_graph_id",
+      "samples": 50,
+      "p50_ms": 0.0970,
+      "p90_ms": 0.1140,
+      "p99_ms": 0.1690,
+      "avg_ms": 0.0886
+    },
+    {
+      "query_id": 7,
+      "name": "learnings_by_user_id",
+      "samples": 50,
+      "p50_ms": 0.0590,
+      "p90_ms": 0.0870,
+      "p99_ms": 9.3100,
+      "avg_ms": 0.2471
+    },
+    {
+      "query_id": 8,
+      "name": "graph_nodes_by_graph_id",
+      "samples": 50,
+      "p50_ms": 0.1100,
+      "p90_ms": 0.1510,
+      "p99_ms": 0.2760,
+      "avg_ms": 0.1104
+    },
+    {
+      "query_id": 9,
+      "name": "project_repos_by_project_id",
+      "samples": 50,
+      "p50_ms": 0.0710,
+      "p90_ms": 0.1030,
+      "p99_ms": 0.1970,
+      "avg_ms": 0.0732
+    },
+    {
+      "query_id": 10,
+      "name": "task_comments_by_task_id",
+      "samples": 50,
+      "p50_ms": 0.1200,
+      "p90_ms": 0.1640,
+      "p99_ms": 0.3300,
+      "avg_ms": 0.1202
+    },
+    {
+      "query_id": 11,
+      "name": "learnings_by_scope",
+      "samples": 50,
+      "p50_ms": 0.0600,
+      "p90_ms": 0.0930,
+      "p99_ms": 0.2700,
+      "avg_ms": 0.0653
+    },
+    {
+      "query_id": 12,
+      "name": "user_settings_by_user_id",
+      "samples": 50,
+      "p50_ms": 0.0490,
+      "p90_ms": 0.0600,
+      "p99_ms": 0.0860,
+      "avg_ms": 0.0471
+    },
+    {
+      "query_id": 13,
+      "name": "workflow_nodes_by_name",
+      "samples": 50,
+      "p50_ms": 0.0420,
+      "p90_ms": 0.0680,
+      "p99_ms": 0.0960,
+      "avg_ms": 0.0434
+    },
+    {
+      "query_id": 14,
+      "name": "graph_node_by_id",
+      "samples": 50,
+      "p50_ms": 0.0800,
+      "p90_ms": 0.1010,
+      "p99_ms": 0.1370,
+      "avg_ms": 0.0783
+    },
+    {
+      "query_id": 15,
+      "name": "user_secrets_by_user_id",
+      "samples": 50,
+      "p50_ms": 0.0470,
+      "p90_ms": 0.0610,
+      "p99_ms": 0.0920,
+      "avg_ms": 0.0472
+    }
+  ]
+}

package/cloud-runtime/standalone/docs/baseline/restore-time.json

@@ -0,0 +1,10 @@
+{
+  "benchmark": "pg_dump_restore",
+  "source_db": "agx",
+  "timestamp": "2026-02-21T22:09:54Z",
+  "dump_size_mb": 41,
+  "dump_seconds": 5.986167000,
+  "restore_seconds": 13.249342000,
+  "pg_host": "localhost",
+  "pg_port": "55432"
+}