From 8eebb99b5301de693764eb8d30e8af8c512fbfeb Mon Sep 17 00:00:00 2001 From: Yoan Blanc Date: Wed, 18 Sep 2024 14:55:26 +0200 Subject: [PATCH] Fix README typos for adaptive sampling (#6000) --- plugin/sampling/strategyprovider/adaptive/README.md | 4 ++-- 1 file changed, 2 insertions(+), 2 deletions(-) diff --git a/plugin/sampling/strategyprovider/adaptive/README.md b/plugin/sampling/strategyprovider/adaptive/README.md index f42f8222646..cf7b762aea2 100644 --- a/plugin/sampling/strategyprovider/adaptive/README.md +++ b/plugin/sampling/strategyprovider/adaptive/README.md @@ -20,9 +20,9 @@ There are three main components of the Adaptive Sampling: Aggregator, Post-aggre ### Post-aggregator -*Post-aggregator* is the main logic responcible for _adaptive_ part of this sampling strategy implementation. Its main job is to load all throughput from storage (because multiple instances of collector could've written different aggregates), aggregate it into a final output, and compute the desired sampling probabilities, which are also written into storage. +*Post-aggregator* is the main logic responsible for _adaptive_ part of this sampling strategy implementation. Its main job is to load all throughput from storage (because multiple instances of collector could've written different aggregates), aggregate it into a final output, and compute the desired sampling probabilities, which are also written into storage. -In a typical production usage Jaeger deployment consists of many collectors. Each collector runs an independent aggregator, because they do not require coordination as long as there is a shared storage. Each collector also runs post-aggregator, however only one of those should be combining the output of all aggregators and producing the final sampling probabilities. This is achived by using a simple leader-follower election with the help of the storage backend. The leader post-aggregator does the main job of the computation, while the follower-aggregators are only loading the throughput from storage and aggregate it in memory, so that each of them is ready to assume the role of the leader if needed, but they do not compute the probabilities or write them back into storage. +In a typical production usage Jaeger deployment consists of many collectors. Each collector runs an independent aggregator, because they do not require coordination as long as there is a shared storage. Each collector also runs post-aggregator, however only one of those should be combining the output of all aggregators and producing the final sampling probabilities. This is achieved by using a simple leader-follower election with the help of the storage backend. The leader post-aggregator does the main job of the computation, while the follower-aggregators are only loading the throughput from storage and aggregate it in memory, so that each of them is ready to assume the role of the leader if needed, but they do not compute the probabilities or write them back into storage. ### Provider