We’ve pushed a round of updates to the doc site based on discussions from the recent open source working group meeting with folks from Apple, Palantir, OpenAI, and Adobe. Here’s what’s new:
Backup V3 — documented the upcoming parallel log upload/download work that Apple is developing for faster restores. Also added notes on multi-cloud backup status (GCS support coming via Palantir, Azure currently broken).
Storage engines — stronger guidance on migrating from SQLite (ssd-2) to Redwood, and updated RocksDB status to reflect real-world production usage at Apple.
Monitoring — added a section on native OpenTelemetry metrics (in progress) and the current workaround of scraping trace events.
Configuration — documented the causal_read_risky knob for GRV path performance.
Testing & benchmarking — added Joshua cluster recommendations for faster test feedback, and documented Mako as a stable benchmarking tool.
Community — welcome OpenAI and Palantir to the production users list!
Very happy to see this, the site looks fantastic! I hope this will make FoundationDB more popular (it’s slightly worrying that it doesn’t see more adoption).
One quick observation: perhaps it might be possible to link to Consistency Models which is I think the best resource on the Internet today listing consistency models with precise explanations. I cringed a bit when I saw “ACID” on the front page, luckily the Guarantees page goes into a lot more detail.
FoundationDB has nothing to be ashamed of here: Strict Serializable level is very rare.
Thanks Jan, great to have you back! You’re absolutely right — “ACID” was underselling what FDB actually provides. We’ve updated the site to properly highlight strict serializable isolation throughout (landing page, ACID guarantees page, and transactions page), and added a link to the Jepsen consistency hierarchy as a reference.
Would love your feedback on the changes — here’s the PR and preview:
Hey all, another round of doc updates based on the working group meeting notes with Apple, Palantir, OpenAI, and Adobe. This one covers some important items we missed in the first pass:
Feature deprecations — Apple has deprecated and removed several features: Dynamic Knobs, Encryption at Rest, Metacluster, Blob Granules, and Parallel Restore. If you’re using or planning to use any of these, check the versions page for details.
Tracing knobs — Documented three recommended knobs (knob_latency_metrics_logging_interval, knob_histogram_report_interval, knob_kaio_latency_logging_interval) set to 60s for more granular performance visibility at minimal cost.
fdb-otel-exporter — Added a reference to the fdb-otel-exporter project as a starting point if you’re looking to export OTel metrics from FDB.
Cluster health metric — Documented the upcoming stable health metric (single 0-100 scalar) that’ll make fleet-wide monitoring much easier. Has an open PR upstream.
Kubernetes — Noted that the fdb-kubernetes-operator is the recommended approach for deploying FDB on K8s.
DR Failover says to run clear range over all keys to break replication. And to run this command on the DR cluster? This might need more explanation as it looks like you are saying to delete all the data on DR after the primary cluster failed. Backup & Recovery - FoundationDB Documentation