deno-memory-profiler

name	deno-memory-profiler
description	Analyze memory in a running Deno process via the V8 inspector CDP protocol. Use when the user asks about memory usage, memory leaks, heap analysis, heap snapshots, allocation profiling, or object retention in a Deno process running with --inspect.

Deno Memory Profiler

Analyze memory in a running Deno process via the V8 inspector CDP protocol.

When to use

Use this skill when the user asks about:

• Memory usage of a running Deno process
• Memory leaks or growing heap
• Heap analysis or heap snapshots
• Allocation profiling
• Object retention or constructor counts

The target Deno process must be running with --inspect or --inspect-brk. The default inspector port is 9229.

Prerequisites

The target process must be started with the inspect flag:

deno run --inspect main.ts          # default port 9229
deno run --inspect=0.0.0.0:9230 main.ts  # custom host/port

Available commands

The script is at skills/deno-memory-profiler/scripts/memory.ts. When run_skill_script is available and exactly allowlisted, invoke it with skill="deno-memory-profiler" and path="scripts/memory.ts" instead of constructing a shell command.

All commands accept --port=<port> (default 9229) and --host=<host> (default 127.0.0.1). Output is JSON to stdout; status messages go to stderr.

usage — Quick heap stats

deno run --allow-net skills/deno-memory-profiler/scripts/memory.ts usage
deno run --allow-net skills/deno-memory-profiler/scripts/memory.ts usage --gc

Returns usedSize, totalSize, usagePercent. Use --gc to force garbage collection before measuring for a more accurate picture.

eval <expression> — Evaluate in target process

deno run --allow-net skills/deno-memory-profiler/scripts/memory.ts eval "Deno.memoryUsage()"
deno run --allow-net skills/deno-memory-profiler/scripts/memory.ts eval "globalThis.myCache.size"

Runs an expression in the target via Runtime.evaluate and prints the result. Useful for inspecting specific objects.

sample — Allocation sampling

deno run --allow-net skills/deno-memory-profiler/scripts/memory.ts sample --duration 10
deno run --allow-net skills/deno-memory-profiler/scripts/memory.ts sample --duration 5 --top 50 --interval 16384

Profiles allocations for the given duration. Returns the top allocation sites sorted by bytes allocated. Options:

• --duration <seconds> — sampling time (default 5)
• --top <N> — number of sites to show (default 30)
• --interval <bytes> — sampling interval (default 32768; lower = more detail, more overhead)

snapshot — Full heap snapshot summary

deno run --allow-net skills/deno-memory-profiler/scripts/memory.ts snapshot

Takes a V8 heap snapshot and produces a summary: total size, top 20 constructors by shallow size, top 10 by instance count, and duplicate strings (>50 occurrences).

diff baseline / diff compare — Leak detection

# Phase 1: capture baseline
deno run --allow-net --allow-write skills/deno-memory-profiler/scripts/memory.ts diff baseline

# ... user triggers suspected leaky operation ...

# Phase 2: compare
deno run --allow-net --allow-read skills/deno-memory-profiler/scripts/memory.ts diff compare

Two-phase leak detection. Baseline saves a snapshot summary to /tmp/memory-baseline-{port}.json. Compare takes a new snapshot and reports:

• Total heap growth (bytes and percent)
• New constructors not in baseline
• Constructors with increased instance count or size
• Constructors that disappeared

Workflow examples

Quick health check

deno run --allow-net skills/deno-memory-profiler/scripts/memory.ts usage --gc

Get the current heap state after GC. Good first step to see if memory is unexpectedly high.

Allocation profiling

deno run --allow-net skills/deno-memory-profiler/scripts/memory.ts sample --duration 10

Run this while the user triggers their workload. Shows where allocations are happening — useful for finding hot allocation paths.

Leak detection

# 1. Take baseline while idle
deno run --allow-net --allow-write skills/deno-memory-profiler/scripts/memory.ts diff baseline

# 2. Ask user to trigger the suspected leaky operation several times

# 3. Compare
deno run --allow-net --allow-read skills/deno-memory-profiler/scripts/memory.ts diff compare

The diff shows what grew. Focus on constructors with large deltaCount or deltaSize.

Deep inspection

# Full snapshot breakdown
deno run --allow-net skills/deno-memory-profiler/scripts/memory.ts snapshot

# Then poke at specific objects
deno run --allow-net skills/deno-memory-profiler/scripts/memory.ts eval "Deno.memoryUsage()"
deno run --allow-net skills/deno-memory-profiler/scripts/memory.ts eval "globalThis.myMap?.size"

Interpreting results

• Large shallow size with high count: Usually arrays, buffers, or string data being retained. Look at what constructor owns them and trace back to why they're kept alive.
• Growing constructors between baseline and compare: These are likely leaks — objects being created but not released. Focus on the ones with the highest deltaCount.
• High allocation count in sampling: Hot allocation paths that may benefit from object pooling, caching, or restructuring to reduce GC pressure.
• Duplicate strings: Many identical strings suggest opportunities for string interning or indicates a pattern creating redundant string copies.
• New constructors in diff compare: Types that didn't exist at baseline may indicate new subsystems being loaded, or leaked closures/objects from the triggered operation.

Limitations

• Snapshot size: Snapshot parsing can be slow and memory-intensive for very large heaps (>500MB). The profiler process itself needs memory to parse the JSON.
• Sampling is duration-based: Coordinate with the user on timing — the sampling period should overlap with the workload being profiled.
• No retained size: Computing retained size requires building a dominator tree, which is expensive. Shallow size grouped by constructor is provided instead, which is still very useful for identifying leaks and large allocations.
• Single target: Connects to the first debuggable target. If multiple isolates are running, only the first is profiled.
• Inspector must be enabled: The target process must be running with --inspect. There's no way to attach to a process that wasn't started with this flag.