1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
|
# Querying ior Parquet recordings with ClickHouse
ior can record I/O events to a Parquet file (see `--record` flag). This document explains
how to explore those recordings interactively using `clickhouse local`, which is bundled
inside the standard `clickhouse/clickhouse-server` Docker image — no server, no persistent
state, no installation needed beyond Docker.
---
## Prerequisites
- Docker running locally (`docker info` must succeed)
- A recording file, e.g. `ior-recording-20260313-170234.parquet`
---
## Schema
| Column | Type | Description |
|---|---|---|
| `seq` | UInt64 | Monotonically increasing event sequence number |
| `time_ns` | UInt64 | Absolute timestamp (nanoseconds since boot) |
| `gap_ns` | UInt64 | Time since previous event (ns) |
| `latency_ns` | UInt64 | Syscall duration (ns) |
| `comm` | String | Process name |
| `pid` | UInt32 | Process ID |
| `tid` | UInt32 | Thread ID |
| `syscall` | String | Syscall name (e.g. `read`, `openat`) |
| `fd` | Int32 | File descriptor |
| `ret` | Int64 | Return value (negative = errno) |
| `bytes` | UInt64 | Bytes transferred (0 if not applicable) |
| `file` | String | File path (empty if not resolved) |
| `is_error` | Bool | True when `ret` is a negative errno |
| `filter_epoch` | UInt64 | Filter generation at capture time |
---
## Running a query
The general pattern mounts the directory containing the file as `/data` inside the container:
```sh
docker run --rm \
-v /path/to/recording/dir:/data:ro \
clickhouse/clickhouse-server:latest \
clickhouse local -q "SELECT ... FROM file('/data/recording.parquet', Parquet)"
```
For convenience, set shell variables:
```sh
FILE="ior-recording-20260313-170234.parquet"
DIR="$(pwd)" # or the absolute directory containing the file
IMG="clickhouse/clickhouse-server:latest"
docker run --rm -v "$DIR:/data:ro" "$IMG" clickhouse local -q \
"SELECT ... FROM file('/data/$FILE', Parquet)"
```
---
## Example queries
### Inspect the schema
```sql
DESCRIBE TABLE file('/data/recording.parquet', Parquet)
```
```
seq UInt64
time_ns UInt64
gap_ns UInt64
latency_ns UInt64
comm String
pid UInt32
tid UInt32
syscall String
fd Int32
ret Int64
bytes UInt64
file String
is_error Bool
filter_epoch UInt64
```
### Row count
```sql
SELECT count(*) FROM file('/data/recording.parquet', Parquet)
```
### Top syscalls by call count
```sql
SELECT syscall, count(*) AS n
FROM file('/data/recording.parquet', Parquet)
GROUP BY syscall
ORDER BY n DESC
LIMIT 15
```
```
read 61800
ioctl 14642
statx 10916
close 8660
openat 6310
write 6053
...
```
### Error breakdown
```sql
SELECT syscall, ret, count(*) AS n
FROM file('/data/recording.parquet', Parquet)
WHERE is_error
GROUP BY syscall, ret
ORDER BY n DESC
LIMIT 15
```
```
read -11 23597 -- EAGAIN (non-blocking, normal)
statx -2 1216 -- ENOENT
ioctl -25 540 -- ENOTTY
openat -2 376 -- ENOENT
...
```
### Latency percentiles per syscall
```sql
SELECT
syscall,
count(*) AS n,
quantile(0.50)(latency_ns) AS p50_ns,
quantile(0.90)(latency_ns) AS p90_ns,
quantile(0.99)(latency_ns) AS p99_ns
FROM file('/data/recording.parquet', Parquet)
GROUP BY syscall
ORDER BY p99_ns DESC
LIMIT 15
```
```
fdatasync 11 1745501 9444892 9994682 -- ~10ms p99, flushes to disk
fallocate 31 35062 487162 589146
rename 3 126619 170921 180889
ftruncate 59 5449 35776 100399
ioctl 14642 1155 8845 63626
...
```
### Top files by I/O bytes
```sql
SELECT file, sum(bytes) AS total_bytes, count(*) AS ops
FROM file('/data/recording.parquet', Parquet)
WHERE file != ''
GROUP BY file
ORDER BY total_bytes DESC
LIMIT 15
```
```
/home/paul/.mozilla/firefox/.../cookies.sqlite-wal 7082880 432
/dev/ptmx 1590757 1472
/proc/3680458/smaps 1555387 419
...
```
### Activity by process
```sql
SELECT comm, pid, count(*) AS n, sum(bytes) AS total_bytes
FROM file('/data/recording.parquet', Parquet)
GROUP BY comm, pid
ORDER BY n DESC
LIMIT 15
```
### Slow syscalls (above threshold)
```sql
SELECT time_ns, comm, pid, syscall, file, latency_ns
FROM file('/data/recording.parquet', Parquet)
WHERE latency_ns > 1000000 -- 1ms
ORDER BY latency_ns DESC
LIMIT 20
```
### Event timeline (10ms buckets)
```sql
SELECT
intDiv(time_ns, 10000000) AS bucket_10ms,
count(*) AS events,
sum(bytes) AS bytes
FROM file('/data/recording.parquet', Parquet)
GROUP BY bucket_10ms
ORDER BY bucket_10ms
```
---
## Automated validation
The `mage parquetValidate` target runs schema, row-count, and sanity checks
against the latest `*.parquet` in the repo root:
```sh
env GOTOOLCHAIN=auto mage parquetValidate
# Or against a specific file:
PARQUET_FILE=ior-recording-20260313-170234.parquet env GOTOOLCHAIN=auto mage parquetValidate
```
It checks:
1. All 14 expected columns are present
2. Row count > 0
3. `seq` is monotonically ordered and `time_ns` is non-zero
|
