V8 8.0: Startup Snapshots for Low‑Traffic Services – Reducing Launch Latency

Answer

Optimal snapshot size: Keep the snapshot small enough that the cost of mapping it into memory (a few milliseconds) is outweighed by the parsing time saved for the bootstrap code. In practice a size under ~5 MB works for most low‑traffic services; larger snapshots increase startup latency linearly.

Effect on minor GC: Restoring a snapshot raises the initial heap size but does not change the GC algorithm. Minor GCs occur when the heap reaches its usual threshold; a larger restored heap may cause the first GC to happen sooner, adding a small latency, but the frequency of GCs during idle periods remains unchanged.

Ignition vs. TurboFan: Code from a snapshot starts in the Ignition interpreter. Functions called only a few times per deployment rarely reach the hot‑threshold needed for TurboFan optimization, so each invocation incurs the higher interpreter cost. If the workload varies, the snapshot may not provide a long‑term CPU benefit for those infrequent functions.

Confirmed Facts

• V8 startup snapshots serialize the heap state at snapshot time, allowing the engine to skip parsing and compilation of the captured code.
• Code executed from a snapshot begins in the Ignition interpreter; TurboFan optimization occurs only after a function exceeds the hot‑threshold.
• Post‑snapshot module loading requires normal parsing and compilation, adding linear latency for any code not included in the snapshot.
• The size of the snapshot is directly proportional to the amount of initialized data and pre‑compiled code stored in the heap at snapshot creation.

Likely Explanation

• A very small snapshot (<5 MB) usually captures only the essential bootstrap, keeping mapping cost low while still saving parsing time for that code.
• Because low‑traffic services allocate little additional heap after startup, the impact on minor GC frequency is minimal; the first GC may occur slightly earlier due to the larger initial heap.
• Functions executed only a handful of times stay in Ignition, so their per‑call CPU usage is higher than it would be after TurboFan optimization in a long‑running isolate.

Steps to Balance Snapshot Size

1. Identify the minimum set of modules required to reach a “ready” state (entry point and core dependencies).
2. Generate a snapshot containing only those modules to keep the binary size manageable.
3. Measure startup time with and without the snapshot using --snapshot-blob and compare; use --print-startup-data to verify the snapshot composition.
4. If post‑snapshot module loading represents a large fraction of the workload, consider asking for the average number of unique modules loaded after restore to refine the target size.