`extract --strategy=complete_ways` is slow with multi-extract configs on large files

[yuiseki]

What version of osmium-tool are you using?

osmium-tool 1.19.1 (also reproduced on current master)

What operating system version are you using?

Ubuntu 24.04, gcc 13, x86_64

Tell us something about your system

32 CPU cores, 94 GB RAM, software RAID array

What did you do exactly?

osmium extract --strategy=complete_ways -c tiles_z2.json planet-latest.osm.pbf

The config file defines 16 bounding boxes covering the full planet at z=2 Web Mercator tile boundaries.

What did you expect to happen?

I expected the runtime not to be dominated by avoidable I/O stalls.

What did happen instead?

Strategy	Elapsed
complete_ways	51m 36s
simple	31m 27s

During Pass 2, disk utilisation reached 100% while CPU was mostly idle, suggesting the main thread was spending much of its time blocked waiting for writes to complete.

What did you do to try analyzing the problem?

I identified two likely causes:

Synchronous write flushing. Extract::write() calls osmium::io::Writer synchronously each time its buffer fills, blocking the main thread while the write is performed.

Redundant way node scanning in Pass 1. Pass1::eway() is called independently for each extract on every way, causing way.nodes() to be scanned multiple times per way.

I implemented fixes for both in my fork and measured on the same planet file and config:

Version	Elapsed	vs baseline
upstream	51m 36s	—
patched	41m 18s	-20%

I plan to submit a pull request. Mentioning here first in case the approach raises concerns.