Sometimes the bitter PIL,
then a comfortable Pillow.
Wrote on read against our will.Now rest in peace —
Thumbor takes it from here.
We're upgrading a Plone 5.2 site to 6.2 right now. The site has tons of images — 230 GB of blobs. I'm running it on the new-stack with zodb-pgjsonb and plone-pgcatalog.
The PITA about Pillow in Plone: every image scale loads the full original blob into memory, resizes it with Pillow, and writes the result back into ZODB. Each scale seems to touch 3 small objects in the database. Multiply that by every image, every scale size — and you understand why we have 230 GB blobs: it produces so much database churn. The data never sleeps, because Pillow keeps waking it up.
So we cut Pillow out of Plone entirely.
plone-pgthumbor replaces image URLs and metadata with direct links and @@images scaling with a 302 redirect to Thumbor. No more Pillow in the Plone process, no more write-on-read. Thumbor does the scaling on the fly, supports smart focal point detection, and caches the results itself. Plone never touches the image bytes. Security is layered: all Thumbor URLs are HMAC-signed so nobody can request arbitrary transformations, and for non-public content the thumborblobloader's auth handler calls back to Plone's @thumbor-auth REST service — a single PostgreSQL query against the allowedRolesAndUsers index, no ZODB object loading needed.
But Thumbor needs to read the original blob. That's where zodb-pgjsonb-thumborblobloader comes in — a Thumbor 7 loader that reads blobs directly from PostgreSQL's blob_state table (or S3 if you tiered them out). Async connection pool, local disk LRU cache, no ZODB, no unpickling. Just bytes from PG to Thumbor.
And with plone-pgcatalog the catalog no longer chokes on all those scale annotations either. PostgreSQL handles the queries — no BTree bloat from thousands of scale entries that nobody ever loads anymore.
This is all early code, do not use yet in production, except if you know what you're doing!
Try it out
Want to see it in action? There's a Docker Compose tryout that gets you a full stack (Plone 6.2 + Thumbor + PostgreSQL + nginx) in minutes. Upload an image, right-click it, and inspect the URL — you'll see the signed Thumbor redirect in action.
Full documentation: plone.pgthumbor 0.1.0 documentation
The before and after
| Old chain | New chain | |
|---|---|---|
| request | request | |
| Plone | Thumbor | |
| ZODB load | PG / S3 | |
| Pillow resize | transform | |
| ZODB store | cache | |
| serve | serve |
Plone just creates a correct stable URL or redirects.
The image data finally sleeps without a Pillow.
