Lightning Talk: Transactional Remote-write and Why You Should Care
summary
Prometheus remote_write only guarantees ordering per series, causing non-atomic commits that break multi-series types like histograms and produce incomplete query results; Timescale/Promscale implements a fix that piggyback on TSDB transactions and adds transaction metadata so the remote can send all samples of a scrape in a single request (or split with start/end markers) and commit atomically. Implementation is in review; design tradeoffs focus on whether grouping by scrape reduces throughput, and preliminary benchmarks are reported.
AI-written summary · about
chapters (4)
related talks
06mLightning Talk
28mIntegrating Long-Term Storage with Prometheus [A]
37mUnlocking Cost Savings & New Possibilities: Your Guide to Prometheus Remote Write 2.0
21mThe Future is Bright, the Future is Prometheus Remote Write
05mLightning Talk: Scrape Trolley Dilemma
47mBeing transactional in 2019 - what does this actually mean?
28mShipping Metrics From the Edge
13mLightning Talk: A Drop-in System To Accelerate Metrics Observability by 100x Us...