Scene 10a

Two scrapers, one copy of the truth — replica labels and deduplication at write or query

A redundant scraper pair sends every sample twice under different replica labels, so the cluster must keep one stream or every sum doubles — either by accepting only an elected replica and paying a short failover gap, or by storing both and merging at query time for double the storage.

Previously

Now that the cluster keeps three safe copies of whatever it receives, the next question is what happens when two redundant scrapers hand it every sample twice.

Scene 10a

Two scrapers, one copy of the truth

Watch
Try it
Predict
Capture

What to watch for

How do you run redundant scrapers without counting everything twice? One scraper is a single point of failure, so teams run two of them side by side: same configuration, same targets, same interval. Both push everything they collect into the cluster, and the cluster from the last scene faithfully stores whatever it is handed. Watch the fleet requests-per-second line fill in against the dashed line of the traffic that is really happening.

Across the top is the six-stage pipeline strip, with store drawn as the cluster.
On the left are two scrapers — a redundant pair collecting the same checkout pods on the same interval, so that losing one machine doesn't blind you.
Each of them pushes its samples into the blue distributor, the cluster's front door from the previous scene.
Every copy carries the same metric and labels plus one extra tag naming which scraper sent it: that tag is the replica label.
The box to the right of the distributor is the cluster's storage, showing how many streams of the same series it is holding (1× or 2×), and the last box is the query side.
Keeping exactly one copy of a series that arrived twice is called deduplication, and the diagram lets you do it at the front door or at the query.
A ✓ elected badge marks the scraper the front door is currently keeping, and the ring beside the pair is the failover timeout — how long the front door waits before electing the other one.
A “202 — dropped” chip is the front door accepting the losing copy and throwing it away.
At the bottom, the fleet requests-per-second chart draws the real traffic as a dashed line and what the graph actually shows as the solid line, and a grey band across it is the stretch where no samples were kept.
Top-right is time-to-detect: how long checkout can be broken before a human is paged.

Sources

Where this sits in Metrics / Monitoring System

Scene 10a of 18, in the Scale out act — Remote-write, ingesters, dedup, blocks, split queries.. A redundant scraper pair sends every sample twice, so the cluster keeps one stream: elect one replica at write time and accept a short failover gap, or store both and merge at query time for twice the storage.

Up next. Now that exactly one stream per series lands in memory, the next question is how the cluster keeps months of it without holding it all in memory.

All 18 scenes in Metrics / Monitoring System · Every curriculum

Built with Arqly

Every scene in Metrics / Monitoring System builds on the one before it.All 18 Metrics / Monitoring System scenes