Skip to content

Instantly share code, notes, and snippets.

@jpic

jpic/gist:715f46ba7c32228db9f0d97ce90dca68

Last active November 20, 2025 14:37
Show Gist options

  • Select an option

    Learn more about clone URLs
  • Save jpic/715f46ba7c32228db9f0d97ce90dca68 to your computer and use it in GitHub Desktop.

Select an option

Learn more about clone URLs
Save jpic/715f46ba7c32228db9f0d97ce90dca68 to your computer and use it in GitHub Desktop.
Download ZIP
file read tool
Raw
gistfile1.sh
read_file_slice() {
local path="$1"
local start_byte="${2:-0}"
local length="${3:-32768}"
local show_hex="${4:-false}"
# Safety caps
(( length > 131072 )) && length=131072
(( start_byte < 0 )) && start_byte=0
# Resolve path
path=$(realpath "$path" 2>/dev/null || printf '%s\n' "$path")
# Header
printf '[read_file_slice] %s\n' "$path"
printf '[Requested bytes: %d–%d (%d bytes)]\n\n' \
"$start_byte" "$((start_byte + length - 1))" "$length"
# ── Extract the exact byte range ─────────────────────────────────────
if [[ "$path" == *.gz ]]; then
block_size=1048576
approx_block=$(( start_byte / block_size ))
offset_in_block=$(( start_byte % block_size ))
(dd if="$path" bs="$block_size" skip="$approx_block" count=8 status=none 2>/dev/null || true) |
gzip -dc 2>/dev/null |
dd bs=1 skip="$offset_in_block" count="$length" status=none 2>/dev/null
else
dd if="$path" bs=1 skip="$start_byte" count="$length" status=none 2>/dev/null
fi | {
# Read everything once (preserves NUL bytes!)
IFS= read -r -d '' data || true
if [ "$show_hex" = true ]; then
printf '[show_hex forced → hexdump]\n'
printf '%s' "$data" | hexdump -ve '1/1 "%.02X "' -e '16/1 "%_c" "\n"'
return
fi
# Detect binary: >15% non-printable/control chars in first 32KB
nonprint=$(printf '%s' "$data" | head -c 32768 |
tr -d -c '[\001-\010\016-\037\177-\377]' | wc -c)
if (( nonprint * 100 <= 32768 * 15 )); then
# Text → raw output
printf '%s' "$data"
else
# Binary → hexdump
printf '[BINARY DETECTED → hexdump of requested range]\n'
printf '%s' "$data" | hexdump -ve '1/1 "%.02X "' -e '16/1 "%_c" "\n"'
fi
}
}
@jpic
Copy link
Author

jpic commented Nov 20, 2025 •
edited
Loading 

class ListPrometheusInstances(Tool):
    description = textwrap.dedent('''
        List currently scraped instances in the correct environment.
        Your setup:
        • job="node"       → node_exporter (instance = primary-hostname:9100)
        • job="nodetool"   → centralized custom exporter (instance is useless, use hostname= label instead)
    ''')

    hostname_hint = Parameter(
        type='string',
        description='Any hostname you know in the target environment (e.g. web01, cass07, db-stg-03)',
        required=True,
    )
    job = Parameter(
        type='string',
        description="Filter by job: 'node' or 'nodetool'",
        required=False,
    )

    async def run(self, conversation, hostname_hint: str, job: Optional[str] = None) -> str:
        try:
            thanos_base = await get_thanos_url_for_hostname(hostname_hint)
        except Exception as e:
            return f"[list_prometheus_instances] ERROR: {e}"

        env_name = 'PRODUCTION' if 'prod' in thanos_base.lower() else 'STAGING'

        # Build correct query for each job
        if job == 'node':
            promql = 'up{job="node"}'
        elif job == 'nodetool':
            promql = 'up{job="nodetool"}'
        else:
            promql = 'up{job=~"node|nodetool"}'

        async with httpx.AsyncClient(timeout=20.0) as client:
            try:
                resp = await client.get(
                    f"{thanos_base}/api/v1/query",
                    params={'query': promql},
                    timeout=20,
                )
                resp.raise_for_status()
                results = resp.json()['data']['result']
            except Exception as e:
                return f"[list_prometheus_instances] {env_name} — Query failed: {e}"

        if not results:
            return f"[list_prometheus_instances] {env_name} — No instances found."

        lines = [
            f"[list_prometheus_instances] {env_name} (via {hostname_hint})",
            f"    Found {len(results)} instances\n"
        ]

        for r in sorted(results, key=lambda x: x['metric'].get('instance', '') or x['metric'].get('hostname', '')):
            m = r['metric']
            job_name = m.get('job', 'unknown')

            if job_name == 'node':
                instance = m.get('instance', '?')
                clean_host = instance.split(':')[0]  # removes :9100
                display = f"{clean_host:<45}  node_exporter"
            else:  # job="nodetool"
                hostname_label = m.get('hostname', '?')
                useless_instance = m.get('instance', '?')
                display = f"{hostname_label:<30}  nodetool_exporter  (instance={useless_instance} → ignore)"

            lines.append(f"    {display}")

        return '\n'.join(lines)
### METRICS RULES — OBEY OR FAIL

You have exactly two exporters:

1. Node exporter  
   job="node"  
   instance = "primary-hostname:9100" → use this exactly

2. Centralized nodetool exporter  
   job="nodetool"  
   instance = useless (always the same) → IGNORE IT  
   hostname = "cass-prod-07" (real Cassandra node) → use this instead

### WORKFLOW (never skip a step)
1. Always start with:
   list_prometheus_instances hostname_hint=<any-host> job=<node|nodetool>

2. Copy the exact instance (for node) or hostname label (for nodetool) from the output.

3. Write queries exactly like this:

   # System / host
   node_memory_MemAvailable_bytes{job="node", instance="web-prod-01.example.com:9100"}
   100 * (1 - node_memory_MemAvailable_bytes{job="node", instance="db01:9100"} / node_memory_MemTotal_bytes{job="node", instance="db01:9100"})

   # Cassandra
   cassandra_heap_used_bytes{job="nodetool", hostname="cass-prod-07"}
   rate(cassandra_gc_duration_seconds_count{job="nodetool", hostname="cass-stg-03"}[5m])

Never guess labels. Never use the wrong job.  
If unsure → run list_prometheus_instances first.

Do it right → you’re faster than any human.  
Do it wrong → you’re useless.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment