Add Dask

dask/benchmark.sh

@@ -0,0 +1,13 @@
+#!/bin/bash
+export BENCH_DOWNLOAD_SCRIPT="download-hits-parquet-partitioned"
+export BENCH_DURABLE=no
+# Dask runs Python expressions directly (server eval()s them).
+# queries.sql holds those Python expressions, one per line, so the
+# default BENCH_QUERIES_FILE=queries.sql in lib/benchmark-common.sh
+# picks them up unchanged.
+# Skip the pre-snapshot ./stop+./start cycle: the loaded
+# state lives only in the daemon's process memory (in-process
+# DataFrame, cluster worker heaps) and stopping wipes it. The
+# playground agent reads this and snapshots the running daemon.
+export PLAYGROUND_SKIP_RESTART_BEFORE_SNAPSHOT=yes
+exec ../lib/benchmark-common.sh

dask/check

@@ -0,0 +1,4 @@
+#!/bin/bash
+set -e
+
+curl -sf http://127.0.0.1:8000/health | grep -q '"ok":true'

dask/data-size

@@ -0,0 +1,4 @@
+#!/bin/bash
+set -e
+
+curl -sS http://127.0.0.1:8000/data-size | python3 -c 'import json,sys; print(json.load(sys.stdin)["bytes"])'

dask/install

@@ -0,0 +1,12 @@
+#!/bin/bash
+set -e
+
+sudo apt-get update -y
+sudo apt-get install -y python3-pip python3-venv
+
+if [ ! -d myenv ]; then
+    python3 -m venv myenv
+fi
+# shellcheck disable=SC1091
+source myenv/bin/activate
+pip install --quiet "dask[dataframe,distributed]" pyarrow fastapi uvicorn

dask/load

@@ -0,0 +1,25 @@
+#!/bin/bash
+set -e
+
+# See duckdb-memory/load for the rationale — the old one-liner
+#     elapsed=$(curl -sS ... | python3 -c '...elapsed...')
+# masked curl/JSON failures with `set -e` and produced a successful-looking
+# zero-timing run when the server died mid-ingest. Capture body and
+# diagnose explicitly.
+body=$(curl -sS -X POST http://127.0.0.1:8000/load 2>&1) || {
+    echo "load: curl to /load failed:" >&2
+    printf '%s\n' "$body" >&2
+    echo "load: server may have been OOM-killed during ingest" >&2
+    exit 1
+}
+
+elapsed=$(printf '%s' "$body" | python3 -c \
+    'import json,sys;print(json.load(sys.stdin)["elapsed"])' 2>&1) || {
+    echo "load: /load did not return valid {\"elapsed\": ...} JSON:" >&2
+    printf '%s\n' "$body" >&2
+    exit 1
+}
+
+echo "Load (server-reported): ${elapsed}s"
+
+sync

dask/queries.sql

@@ -0,0 +1,43 @@
+hits.count()
+hits[hits['AdvEngineID'] != 0].count()
+(hits['AdvEngineID'].sum(), hits.shape[0], hits['ResolutionWidth'].mean())
+hits['UserID'].mean()
+hits['UserID'].nunique()
+hits['SearchPhrase'].nunique()
+(hits['EventDate'].min(), hits['EventDate'].max())
+hits[hits['AdvEngineID'] != 0].groupby('AdvEngineID').size().rename('c').reset_index().sort_values('c', ascending=False)
+hits.groupby('RegionID')['UserID'].nunique().nlargest(10)
+hits.groupby('RegionID').agg(AdvEngineID=('AdvEngineID', 'sum'), c=('WatchID', 'size'), ResolutionWidth=('ResolutionWidth', 'mean'), UserID=('UserID', nunique)).nlargest(10, 'c')
+hits[hits['MobilePhoneModel'] != ''].groupby('MobilePhoneModel')['UserID'].nunique().nlargest(10)
+hits[hits['MobilePhoneModel'] != ''].groupby(['MobilePhone', 'MobilePhoneModel'])['UserID'].nunique().nlargest(10)
+hits[hits['SearchPhrase'] != ''].groupby('SearchPhrase').size().nlargest(10)
+hits[hits['SearchPhrase'] != ''].groupby('SearchPhrase')['UserID'].nunique().nlargest(10)
+hits[hits['SearchPhrase'] != ''].groupby(['SearchEngineID', 'SearchPhrase']).size().nlargest(10)
+hits.groupby('UserID').size().nlargest(10)
+hits.groupby(['UserID', 'SearchPhrase']).size().nlargest(10)
+hits.groupby(['UserID', 'SearchPhrase']).size().head(10)
+hits.groupby([hits['UserID'], hits['EventTime'].dt.minute, 'SearchPhrase']).size().nlargest(10)
+hits[hits['UserID'] == 435090932899640449]
+hits[hits['URL'].str.contains('google')].shape[0]
+hits[hits['URL'].str.contains('google') & (hits['SearchPhrase'] != '')].groupby('SearchPhrase').agg(URL=('URL', 'min'), c=('SearchPhrase', 'size')).nlargest(10, 'c')
+hits[hits['Title'].str.contains('Google') & ~hits['URL'].str.contains('.google.') & (hits['SearchPhrase'] != '')].groupby('SearchPhrase').agg(URL=('URL', 'min'), Title=('Title', 'min'), c=('SearchPhrase', 'size'), UserID=('UserID', nunique)).nlargest(10, 'c')
+hits[hits['URL'].str.contains('google')].nsmallest(10, 'EventTime')
+hits[hits['SearchPhrase'] != ''].nsmallest(10, 'EventTime').compute()[['SearchPhrase']]
+hits[hits['SearchPhrase'] != ''][['SearchPhrase']].sort_values('SearchPhrase').head(10, npartitions=-1)
+hits[hits['SearchPhrase'] != ''].sort_values(['EventTime', 'SearchPhrase']).head(10, npartitions=-1)[['SearchPhrase']]
+hits[hits['URL'] != ''].assign(l=hits['URL'].str.len()).groupby('CounterID').agg(l=('l', 'mean'), c=('URL', 'size')).query('c > 100000').nlargest(25, 'l')
+hits[hits['Referer'] != ''].assign(k=lambda d: d['Referer'].str.extract('^https?://(?:www\\.)?([^/]+)/.*$')[0], l=lambda d: d['Referer'].str.len()).groupby('k').agg(l=('l', 'mean'), c=('Referer', 'size'), min_referer=('Referer', 'min')).query('c > 100000').nlargest(25, 'l')
+[(hits['ResolutionWidth'] + i).sum() for i in range(90)]
+hits[hits['SearchPhrase'] != ''].groupby(['SearchEngineID', 'ClientIP']).agg(c=('SearchEngineID', 'size'), IsRefreshSum=('IsRefresh', 'sum'), AvgResolutionWidth=('ResolutionWidth', 'mean')).nlargest(10, 'c')
+hits[hits['SearchPhrase'] != ''].groupby(['WatchID', 'ClientIP']).agg(c=('WatchID', 'size'), IsRefreshSum=('IsRefresh', 'sum'), AvgResolutionWidth=('ResolutionWidth', 'mean')).nlargest(10, 'c')
+hits.groupby(['WatchID', 'ClientIP']).agg(c=('WatchID', 'size'), IsRefreshSum=('IsRefresh', 'sum'), AvgResolutionWidth=('ResolutionWidth', 'mean')).nlargest(10, 'c')
+hits.groupby('URL').size().rename('c').nlargest(10).reset_index()
+hits.groupby('URL').size().rename('c').nlargest(10).reset_index()
+hits.assign(**{f'ClientIP_minus_{i}': hits['ClientIP'] - i for i in range(1, 4)}).groupby(['ClientIP', 'ClientIP_minus_1', 'ClientIP_minus_2', 'ClientIP_minus_3']).size().rename('c').nlargest(10).reset_index()
+hits[(hits['CounterID'] == 62) & (hits['EventDate'] >= '2013-07-01') & (hits['EventDate'] <= '2013-07-31') & (hits['DontCountHits'] == 0) & (hits['IsRefresh'] == 0) & (hits['URL'] != '')].groupby('URL').size().nlargest(10)
+hits[(hits['CounterID'] == 62) & (hits['EventDate'] >= '2013-07-01') & (hits['EventDate'] <= '2013-07-31') & (hits['DontCountHits'] == 0) & (hits['IsRefresh'] == 0) & (hits['Title'] != '')].groupby('Title').size().nlargest(10)
+hits[(hits['CounterID'] == 62) & (hits['EventDate'] >= '2013-07-01') & (hits['EventDate'] <= '2013-07-31') & (hits['IsRefresh'] == 0) & (hits['IsLink'] != 0) & (hits['IsDownload'] == 0)].groupby('URL').size().rename('PageViews').nlargest(1010).reset_index().compute().iloc[1000:1010]
+hits[(hits['CounterID'] == 62) & (hits['EventDate'] >= '2013-07-01') & (hits['EventDate'] <= '2013-07-31') & (hits['IsRefresh'] == 0)].assign(Src=lambda d: d['Referer'].where((d['SearchEngineID'] == 0) & (d['AdvEngineID'] == 0), '')).groupby(['TraficSourceID', 'SearchEngineID', 'AdvEngineID', 'Src', 'URL']).size().rename('PageViews').nlargest(1010).reset_index().compute().iloc[1000:1010]
+hits[(hits['CounterID'] == 62) & (hits['EventDate'] >= '2013-07-01') & (hits['EventDate'] <= '2013-07-31') & (hits['IsRefresh'] == 0) & hits['TraficSourceID'].isin([-1, 6]) & (hits['RefererHash'] == 3594120000172545465)].groupby(['URLHash', 'EventDate']).size().rename('PageViews').nlargest(110).reset_index().compute().iloc[100:110]
+hits[(hits['CounterID'] == 62) & (hits['EventDate'] >= '2013-07-01') & (hits['EventDate'] <= '2013-07-31') & (hits['IsRefresh'] == 0) & (hits['DontCountHits'] == 0) & (hits['URLHash'] == 2868770270353813622)].groupby(['WindowClientWidth', 'WindowClientHeight']).size().rename('PageViews').nlargest(10010).reset_index().compute().iloc[10000:10010]
+hits[(hits['CounterID'] == 62) & (hits['EventDate'] >= '2013-07-14') & (hits['EventDate'] <= '2013-07-15') & (hits['IsRefresh'] == 0) & (hits['DontCountHits'] == 0)].assign(M=lambda d: d['EventTime'].dt.floor('min')).groupby('M').size().rename('PageViews').reset_index().compute().sort_values('M').iloc[1000:1010]

dask/query

@@ -0,0 +1,28 @@
+#!/bin/bash
+# Reads a query from stdin, dispatches to the running in-VM server.
+# Stdout: result (rendered table or scalar from the server).
+# Stderr: query runtime in fractional seconds on the last line.
+# Exit non-zero on error.
+set -e
+
+query=$(cat)
+
+tmp=$(mktemp)
+status=$(curl -sS -o "$tmp" -w '%{http_code}' \
+    -X POST --data-binary @- http://127.0.0.1:8000/query <<<"$query")
+
+body=$(cat "$tmp")
+rm -f "$tmp"
+
+if [ "$status" != "200" ]; then
+    echo "query failed: HTTP $status: $body" >&2
+    exit 1
+fi
+
+# Pull `result` for stdout and `elapsed` for stderr (host timing protocol).
+python3 - "$body" <<'PY'
+import json, sys
+d = json.loads(sys.argv[1])
+print(d.get("result", ""))
+sys.stderr.write(str(d["elapsed"]) + "\n")
+PY

dask/results/20260723/c6a.metal.json

@@ -0,0 +1,60 @@
+{
+      "system": "Dask (DataFrame)",
+      "date": "2026-07-23",
+      "machine": "c6a.metal",
+      "cluster_size": 1,
+      "proprietary": "no",
+      "hardware": "cpu",
+      "tuned": "no",
+      "tags": ["Python","dataframe","in-memory","column-oriented","lukewarm-cold-run"],
+      "load_time": 40,
+      "data_size": 174168110180,
+      "concurrent_qps": 0.082,
+      "concurrent_error_ratio": 0.155,
+      "result": [
+        [64.222, 7.563, 7.727],
+        [58.304, 0.71, 0.707],
+        [57.661, 0.409, 0.254],
+        [57.432, 0.278, 0.257],
+        [58.667, 1.227, 1.162],
+        [59.447, 2.335, 2.309],
+        [57.666, 0.256, 0.311],
+        [58.037, 0.719, 0.758],
+        [59.549, 2.382, 2.306],
+        [null, null, null],
+        [67.082, 10.4, 9.805],
+        [68.578, 11.581, 11.438],
+        [61.524, 4.222, 4.296],
+        [67.714, 9.871, 11.01],
+        [113.814, 53.797, 54.739],
+        [59.166, 1.54, 1.588],
+        [142.554, 90.388, 88.26],
+        [64.71, 7.225, 7.211],
+        [136.769, 71.584, 68.834],
+        [57.715, 0.58, 0.514],
+        [null, null, null],
+        [null, null, null],
+        [null, null, null],
+        [null, null, null],
+        [58.351, 0.89, 0.948],
+        [null, null, null],
+        [63.587, 6.332, 6.151],
+        [60.525, 3.193, 3.226],
+        [null, null, null],
+        [61.464, 4.507, 4.705],
+        [71.409, 13.903, 14.178],
+        [209.28, 153.635, 153.406],
+        [208.328, 155.147, 154.055],
+        [69.802, 12.435, 12.374],
+        [69.473, 12.61, 12.722],
+        [76.846, 20.591, 20.255],
+        [59.272, 1.771, 2.214],
+        [59.065, 1.468, 1.418],
+        [58.574, 0.849, 0.917],
+        [61.266, 3.688, 3.947],
+        [58.726, 1.01, 1.018],
+        [58.432, 0.918, 0.903],
+        [58.448, 0.932, 0.923]
+]
+  } 
+

dask/results/20260724/c6a.metal.json

@@ -0,0 +1,60 @@
+{
+      "system": "Dask (DataFrame)",
+      "date": "2026-07-24",
+      "machine": "c6a.metal",
+      "cluster_size": 1,
+      "proprietary": "no",
+      "hardware": "cpu",
+      "tuned": "no",
+      "tags": ["Python","dataframe","in-memory","column-oriented","lukewarm-cold-run"],
+      "load_time": 71,
+      "data_size": 174168110180,
+      "concurrent_qps": 0.08,
+      "concurrent_error_ratio": 0.172,
+      "result": [
+        [64.327, 7.488, 7.706],
+        [58.413, 0.784, 0.911],
+        [57.779, 0.421, 0.251],
+        [57.608, 0.277, 0.276],
+        [58.342, 1.256, 1.233],
+        [59.416, 1.898, 2.104],
+        [57.507, 0.288, 0.287],
+        [57.919, 0.729, 0.703],
+        [60.01, 2.351, 2.709],
+        [null, null, null],
+        [67.568, 10.112, 10.153],
+        [68.678, 11.419, 11.295],
+        [61.52, 4.124, 4.377],
+        [67.667, 10.49, 10.105],
+        [105.971, 52.709, 54.968],
+        [58.969, 1.561, 1.493],
+        [141.452, 96.276, 86.095],
+        [64.34, 7.333, 6.791],
+        [134.602, 71.966, 75.828],
+        [57.882, 0.556, 0.475],
+        [null, null, null],
+        [null, null, null],
+        [null, null, null],
+        [null, null, null],
+        [58.34, 0.908, 0.868],
+        [null, null, null],
+        [63.947, 6.158, 6.867],
+        [61.09, 3.037, 3.529],
+        [null, null, null],
+        [61.51, 4.562, 4.68],
+        [71.334, 14.211, 13.895],
+        [209.84, 152.482, 154.958],
+        [209.852, 154.998, 151.511],
+        [69.724, 11.831, 12.081],
+        [69.544, 12.503, 12.443],
+        [77.534, 19.754, 20.131],
+        [59.624, 1.754, 2.013],
+        [58.879, 1.275, 1.494],
+        [58.487, 0.936, 0.879],
+        [61.248, 3.826, 3.788],
+        [58.826, 1.12, 1.043],
+        [58.374, 1.048, 0.831],
+        [58.59, 1.009, 0.974]
+]
+  } 
+

dask/results/20260724/c7a.metal-48xl.json

@@ -0,0 +1,60 @@
+{
+      "system": "Dask (DataFrame)",
+      "date": "2026-07-24",
+      "machine": "c7a.metal-48xl",
+      "cluster_size": 1,
+      "proprietary": "no",
+      "hardware": "cpu",
+      "tuned": "no",
+      "tags": ["Python","dataframe","in-memory","column-oriented","lukewarm-cold-run"],
+      "load_time": 69,
+      "data_size": 174168110180,
+      "concurrent_qps": 0.092,
+      "concurrent_error_ratio": 0.167,
+      "result": [
+        [64.311, 6.044, 6.273],
+        [58.61, 0.59, 0.584],
+        [57.611, 0.287, 0.244],
+        [57.14, 0.225, 0.216],
+        [58.022, 1.115, 1.018],
+        [59.562, 1.782, 1.591],
+        [57.425, 0.241, 0.22],
+        [57.787, 0.545, 0.563],
+        [58.246, 1.871, 1.906],
+        [null, null, null],
+        [66.509, 9.188, 9.383],
+        [67.914, 10.232, 10.365],
+        [61.034, 3.578, 3.719],
+        [66.957, 9.32, 9.115],
+        [102.899, 47.573, 47.836],
+        [58.648, 1.288, 1.311],
+        [140.049, 91.377, 83.673],
+        [63.275, 6.373, 6.311],
+        [132.172, 71.46, 68.367],
+        [57.783, 0.454, 0.388],
+        [null, null, null],
+        [null, null, null],
+        [null, null, null],
+        [null, null, null],
+        [57.919, 0.823, 0.774],
+        [null, null, null],
+        [62.798, 5.355, 5.517],
+        [59.995, 2.525, 2.717],
+        [null, null, null],
+        [60.473, 3.712, 3.754],
+        [71.95, 14.743, 14.428],
+        [231.807, 174.428, 176.522],
+        [233.801, 173.07, 177.225],
+        [68.069, 11.209, 11.097],
+        [67.907, 11.658, 11.041],
+        [78.273, 20.86, 20.238],
+        [58.939, 1.52, 1.522],
+        [58.652, 1.115, 1.019],
+        [58.262, 0.761, 0.632],
+        [60.146, 3.115, 3.207],
+        [58.084, 0.778, 0.782],
+        [58.284, 0.8, 0.616],
+        [58.039, 0.824, 0.764]
+]
+  } 
+

dask/results/20260724/c8g.metal-48xl.json

@@ -0,0 +1,60 @@
+{
+      "system": "Dask (DataFrame)",
+      "date": "2026-07-24",
+      "machine": "c8g.metal-48xl",
+      "cluster_size": 1,
+      "proprietary": "no",
+      "hardware": "cpu",
+      "tuned": "no",
+      "tags": ["Python","dataframe","in-memory","column-oriented","lukewarm-cold-run"],
+      "load_time": 69,
+      "data_size": 174168110180,
+      "concurrent_qps": 0.108,
+      "concurrent_error_ratio": 0.156,
+      "result": [
+        [62.696, 4.739, 4.583],
+        [58.581, 0.485, 0.484],
+        [58.325, 0.204, 0.193],
+        [56.962, 0.143, 0.165],
+        [57.693, 0.798, 0.808],
+        [58.185, 1.346, 1.292],
+        [56.986, 0.135, 0.136],
+        [57.324, 0.426, 0.424],
+        [58.487, 1.599, 1.616],
+        [null, null, null],
+        [63.786, 6.839, 6.854],
+        [64.769, 7.835, 7.924],
+        [60.103, 3.324, 3.296],
+        [63.45, 6.64, 6.689],
+        [94.583, 37.84, 37.699],
+        [57.987, 1.127, 1.113],
+        [120.212, 62.556, 62.337],
+        [61.859, 5.143, 5.089],
+        [112.079, 52.355, 53.691],
+        [57.159, 0.382, 0.326],
+        [null, null, null],
+        [null, null, null],
+        [null, null, null],
+        [null, null, null],
+        [57.604, 0.577, 0.581],
+        [null, null, null],
+        [61.807, 4.925, 4.442],
+        [59.267, 2.134, 2.08],
+        [null, null, null],
+        [59.457, 2.794, 2.899],
+        [66.703, 9.836, 9.601],
+        [162.965, 107.044, 106.46],
+        [163.732, 108.041, 107.67],
+        [66.84, 9.45, 9.142],
+        [66.628, 9.691, 9.656],
+        [70.519, 13.682, 13.813],
+        [58.274, 1.192, 1.231],
+        [57.728, 0.773, 0.794],
+        [57.567, 0.571, 0.59],
+        [59.564, 2.546, 2.64],
+        [57.742, 0.658, 0.658],
+        [68.956, 0.713, 0.62],
+        [57.639, 0.586, 0.64]
+]
+  } 
+