Skip to content

BuildNot yet confirmed elsewhere1 publisher2 min readPublished

EVE Online traces a 20-year-old single-shard game with head-based sampling only

EVE Online runs 100% head-based sampling with no tail sampling, Nicholas Herring said in a talk published by InfoQ. The team blends deterministic and exponential-moving-average sampling to balance trace fidelity against cost.

The Engineer · Build desk

How we use AISend a correction

Photograph accompanying EVE Online traces a 20-year-old single-shard game with head-based sampling only
Photo: infoq.com
EVE counts every bullet among its items Item and daily market transaction counts Herring gave for EVE Online, and how much of its sampling is head-based.

Key figures from Nicholas Herring's talk: a 12 billion item count that includes every bullet in EVE Online's universe; about 500,000 market transactions per day; and 100% head-based sampling, with no tail-based sampling.

EVE counts every bullet among its items
FigureValueNoteClaim
Item count, including every bullet in the universe12,000,000,000 items9
Market transactions, about500,000 per day13
Head-based sampling100%2

What happened

  • Herring called exponential moving average sampling, the dynamic half of the blend, a "blessing and curse."
  • EVE runs as a single shard with no regional servers, so every player shares one universe on one server.
  • Shared play forces the game to synchronize, replicate and keep data transactional, a problem Herring compared with FinTech.

Why it matters

  • constraint A rare failure or slow request survives in EVE's traces only if the sampler happened to keep it at the start, because nothing downstream can promote a dropped trace.
  • decision Teams weighing head against tail sampling first have to decide what they must be able to reconstruct. EVE's answer fits traffic that includes player expression the system need not care about.
  • precedent Head-only sampling with a dynamic component now has a named production example, but the evidence is one team's account in one talk.

Head sampling makes its keep-or-drop call when a trace starts, before anything about the outcome exists. A head sampler cannot pick out the slow or failed traces afterwards. Tail sampling decides once the trace is complete. [14]

Herring does not dispute the cost. He said the choice "makes a lot of people clutch their pearls and gasp" and that the team is "buying in a lot of those tradeoffs." [3] The transcript excerpt stops during the scale discussion, before any rate, window or threshold appears and before it says which traffic the deterministic rules cover.

The scale figures set the volume. Twelve billion items across about half a million market transactions is 24,000 items per transaction on average [13]. The average is loose. Herring said the item count includes "literally every single bullet that's in the universe," and that the talk has to deal with tracing "all of the bullets and lasers and missiles flying around in the ecosystem" [9].

He also gave a reason some loss is tolerable. Video games, he said, do "a lot of stuff that the system doesn't really need to care about," and he used waving at another player as the example [11]. In our view that context is what makes 100% head-based sampling defensible here. A dropped trace of a wave costs nobody anything.

The choice transfers only to a system where a dropped trace is cheap to lose. A payment path that has to explain every failure would not qualify. We'd expect the deterministic half of EVE's blend to cover the paths that cannot be dropped.

What to watch

  • The rest of the transcript: whether it states sampling rates, the moving-average window, and how the deterministic rules combine with the dynamic ones.
  • Whether EVE's team reports incidents where a trace dropped at the head would have shortened the diagnosis.
  • Herring's separate talk on running EVE as a single shard, which he said covers the technical challenges of that design.

Clarity's read

What the record supports and how the coverage leans. The claims behind it follow.

Reality

Evidence40
Adoption50
Hype gap+5
Incentives
Insufficient
Confidence55
Why these scores

Claim ledger

Ranked by verification strength, evidence, and original report placement.

  1. [1]

    Nicholas Herring presented how EVE Online adopted observability into a system that is 20 years old and has newer mechanisms evolving underneath it.

    ReportedSupportedSource: Nicholas Herring, InfoQ presentation transcriptView cited source
  2. [2]

    EVE Online's system is 100% head-based sampling and does not use tail-based sampling.

    ReportedSupportedSource: Nicholas Herring, InfoQ presentation transcriptView cited source
  3. [3]

    Herring said the 100% head-based approach "makes a lot of people clutch their pearls and gasp, because we don't use tail base sampling which means we're buying in a lot of those tradeoffs."

    ReportedSupportedSource: Nicholas Herring, InfoQ presentation transcriptView cited source

Sources

1 independent publisher whose own reporting we read for this story.

  1. infoq.com

    1 article · October 13, 2026

    Presentation: How Eve Online Leverages Head Based Sampling to Observe "Fun"

Share your take

Let Clarity write the post for you.

Signed-in readers get a short post drafted on this story in the register they choose — narrative, analytical, or a direct position — editable to the last word before it goes anywhere. The share buttons at the top of this story work without an account.

Topics and entities

Follow any of these and your For You feed starts watching them — no settings page required.

Topics

Entities

Loading related stories