Release notes

Large projects set up faster0.15.2

Imports now work on several parts of a document at once, so a large project is ready sooner. A background job that reprocesses spec books also stops retrying forever when a book is large.

2026-09-22Patch1 change commitNo migrations

Highlights

Imports process several parts of a document at once

Layr splits each uploaded PDF into batches of pages. Since 0.15.0 production worked on one batch at a time while we fixed what broke under load. It now works on up to four at once, so large plan sets and spec books take less time to become ready. On dev, a three-document project of 2,421 pages went from queued to ready in under four minutes with four at once.

Large spec books no longer get stuck in reprocessing

A background job reads a document's pages again to refresh its page names and structure. On a spec book with more than about a thousand pages, that job failed on every try and was retried without limit. The document showed as still processing, and it could not be reprocessed while the job was running. Two books on dev were retried more than 10,000 times each. The job now reads pages in small batches, so large books go through. If it still fails, it stops after a fixed number of tries and records why.

Upgrade notes

No migrations.
Production's import worker count goes from 1 to 4. It is the IMPORT_RANGE_WORKERS key in the Modal layr-pipeline secret on layr-main, set just before this release merged so the release deploy picks it up. Dev already runs at 4.
To roll back the worker count, set IMPORT_RANGE_WORKERS back to 1 in that secret and redeploy layr-pipeline on layr-main. No code change is needed.
A job that was already stuck runs the fixed code on its next try after the deploy. If that try fails, the job stops at once, because it has already used up its tries.

Known limits

Four at once is a cap, not a target. A small document with one batch still runs as one.
When the job fails for good, the document says it couldn’t read the original PDF, whatever the real cause. The reason is in the job’s record, not on screen.