DuckDB

DuckDB reads execution trace Parquet files directly — no server, no import step.

TRACES=~/.grog/*/cache/traces

# Recent builds overview
duckdb -c "
  SELECT trace_id[:8] AS id, command, total_targets, cache_hit_count,
         total_duration_millis / 1000.0 AS sec, git_commit[:7] AS commit
  FROM read_parquet('$TRACES/builds/**/*.parquet')
  ORDER BY start_time_unix_millis DESC LIMIT 20;
"

# Top 10 slowest targets (cache misses only)
duckdb -c "
  SELECT label,
         AVG(command_duration_millis)::INT AS avg_cmd_ms,
         AVG(queue_wait_millis)::INT AS avg_queue_ms,
         COUNT(*) AS n
  FROM read_parquet('$TRACES/spans/**/*.parquet')
  WHERE cache_result = 'CACHE_MISS'
  GROUP BY label HAVING n > 3
  ORDER BY avg_cmd_ms DESC LIMIT 10;
"

# Cache hit rate trend (last 2 weeks)
duckdb -c "
  SELECT DATE_TRUNC('day', EPOCH_MS(start_time_unix_millis)) AS day,
         SUM(cache_hit_count)::FLOAT / SUM(total_targets) * 100 AS hit_pct,
         COUNT(*) AS builds
  FROM read_parquet('$TRACES/builds/**/*.parquet')
  GROUP BY day ORDER BY day DESC LIMIT 14;
"

# Targets where most time is spent in I/O (potential cache bottleneck)
duckdb -c "
  SELECT label,
         AVG(output_write_millis + output_load_millis + cache_write_millis)::INT AS avg_io_ms,
         AVG(command_duration_millis)::INT AS avg_cmd_ms,
         COUNT(*) AS n
  FROM read_parquet('$TRACES/spans/**/*.parquet')
  GROUP BY label ORDER BY avg_io_ms DESC LIMIT 10;
"

S3 and GCS access

For traces stored on S3 or GCS:

# S3
duckdb -c "
  CREATE SECRET (TYPE S3, REGION 'us-east-1');
  SELECT * FROM read_parquet('s3://my-bucket/traces/builds/**/*.parquet')
  ORDER BY start_time_unix_millis DESC LIMIT 10;
"

# GCS
duckdb -c "
  CREATE SECRET (TYPE GCS, KEY_ID 'GOOG...', SECRET '...');
  SELECT * FROM read_parquet('gcs://my-bucket/traces/builds/**/*.parquet')
  ORDER BY start_time_unix_millis DESC LIMIT 10;
"