Running Sentry?

Sentry noisy, over quota, or stack traces useless? We clear the backlog.

Sentry is at its best when it tells you about the five issues that actually matter and ignores the fifty that do not. Most of the Sentry work we pick up is the other way round: a project drowning in false alerts, source maps that never uploaded, and a quota that runs out on the twentieth of the month. We cut the noise and keep the signal.

  • Since 2015Fixing technology
  • FreeEstimates
  • US teamReal engineers

sentry-config.log

still broken

What it is

Sentry at a glance

Sentry is an error monitoring and performance tool first released in 2008 as an open-source Django add-on by David Cramer. It is now maintained by Sentry (functional-software.com) and used by hundreds of thousands of organisations to capture exceptions, performance traces, session replays, and profiling data from web, mobile, and backend applications.

The product covers over 100 SDKs across JavaScript, Python, Ruby, Go, Java, .NET, PHP, iOS, Android, React Native, and more. It offers a hosted SaaS version and a self-hosted option under the Functional Source License (previously BSL). Pricing is based on event, transaction, replay, and profiling volume, with plans from free developer to enterprise, and the scale of ingestion is almost always what causes the invoice conversation.

Type
Error monitoring, performance, session replay, profiling
First released
2008
License
Functional Source License (self-hosted), proprietary SaaS
Maintained by
Sentry (functional-software.com)
Typically used for
Catching exceptions, tracing slow requests, replaying user sessions, alerting on regressions, SDLC integrations with GitHub, Jira, Slack

The honest version

Sentry pros and cons

No technology is all upside. This is where Sentry earns its place and where it tends to cause trouble.

Where Sentry is strong

  • A great developer-first UIIssue pages show the stack trace, breadcrumbs, user, release, environment and a replay, side by side. Debugging a reported bug starts with a URL rather than a search.
  • SDKs for everythingWeb, mobile, server and function-as-a-service runtimes all have first-party SDKs. Attaching Sentry to a new service is minutes of work, not hours.
  • Release tracking and regressionsAssociating issues with a release tells you which deploy introduced a bug, and regression alerts fire when a resolved issue starts happening again.
  • Session replay for frontend bugsReplay rebuilds the user session around an error. For hard-to-reproduce frontend issues, it collapses a one-week ticket into a five-minute fix.
  • Performance and profiling in the same placeSpans, transactions and profiles line up next to the errors they affect. Teams no longer have to pivot between two vendors to answer one question.
  • Good integrationsGitHub, GitLab, Jira, Slack and PagerDuty integrations route issues to the right team without stitching webhooks together.

Where Sentry bites

  • Cost scales with volume, not valueA noisy app can pay for events that no one will ever read. Sampling and inbound filters have to be configured deliberately or the invoice simply follows traffic.
  • Source maps are a persistent painUpload failures, mismatched release names, hashed filenames that do not match, and stack traces minified into unreadable noise. Every frontend team has lost hours to this at least once.
  • Alerting needs real designThe default alert on every new issue fires too often. Useful alerts need issue filters, frequency rules and owners, and most teams only reach that after their inbox gives up.
  • Performance can swamp errorsTransactions are counted and billed separately from events, and enabling tracing on a chatty service can blow a budget overnight if dynamic sampling is not tuned.
  • Self-hosted is a real operationThe self-hosted stack is Docker Compose with roughly 20 services including Snuba, ClickHouse, Kafka and Redis. It runs, and it takes real ops time.
  • Data retention is tier-dependentHistorical retention varies by plan, and a long-tail bug can time out of the window before the team gets to it, which leaves the issue reopen with no history.

Sound familiar?

Common Sentry problems we fix

The Sentry problems that bring people to us, and what you have once each one is fixed.

4.2M of 5M events

Sentry is burning through the plan early

A single page throwing 20 errors per visit, a background job reporting every retry as a new event, or a library emitting ResizeObserver warnings on every scroll. The noise is where the budget goes.

Noise cut, signal preserved

source maps failed to upload

Stack traces are unreadable

Hashed filenames at build, a sentry-cli upload step that silently returned a 403, or a release name that drifted between CI and the SDK. Minified traces land in the inbox and no one can act on them.

Traces mapped back to source

Non-Error promise rejection

Non-error rejections flood the inbox

A promise that rejects with a string, an abort from the user navigating away, or a fetch that threw a plain object instead of an Error. Sentry groups these oddly and they pile up.

Only real errors in the queue

no alert ownership

Alerts fire and nobody owns them

A project set up years ago with an alert on every new issue going to a mailing list nobody reads. Pages get muted, and real regressions sit next to noise.

Alerts owned, routed, acted on

sample rate: 100%

Tracing cost ballooned overnight

Someone enabled full-rate tracing on a service with a million requests a day, forgetting dynamic sampling. The next invoice arrives with the bad news.

Sampling honest, cost predictable

release not found

Releases are not being created

The CI step that creates the release is failing or missing, so events arrive tagged with releases Sentry has no record of, and regression detection does not work.

Releases created on every deploy

duplicate issues: 412

The same bug is captured as hundreds of issues

A fingerprint rule that is too strict, or a dynamic message that changes the grouping key on every event. The dashboard looks like fifty new bugs when it is really one.

Fingerprints stable, grouping sane

self-hosted: 24 containers

The self-hosted stack is drifting

A self-hosted Sentry two major versions behind, Snuba or ClickHouse out of sync, and the last upgrade attempted during a Friday afternoon that nobody wants to repeat.

Self-hosted upgraded or lifted to SaaS

What we do

Sentry help, from rescue to upkeep

Pick up the phone for one noisy project, or hand us the Sentry org across all your services.

Rescue
Inbox out of control, quota hit mid-month, stack traces useless. We get the noise down and the signal visible again fast.
Quota and sampling
Inbound filters, dynamic sampling, release tags, performance budgets, with the monthly event count and invoice before and after.
Source maps and releases
sentry-cli integrated into the CI pipeline, release names consistent, source maps uploaded and verified against real stack traces.
Alert design
Issue owners, frequency rules, escalation to PagerDuty for the ones that matter, and silence on the ones that do not.
Self-hosted upkeep
Upgrades, ClickHouse and Kafka health, backups, and a recommendation on whether SaaS is cheaper than the hours self-hosted keeps costing.
Managed Tech
A monthly look at event volume, alert quality and open issue age, plus an engineer on hand when Sentry itself wobbles alongside the rest of the stack.

How it works

From first message to fixed

  1. Look Free estimate

    Tell us what's going wrong. We look at it, then tell you plainly what's wrong and what it will cost.

  2. Fix

    We fix the cause, not the symptom. If something is down, getting it back up comes first.

  3. Harden

    Backups, monitoring, updates and access, so the same problem can't quietly come back.

  4. Hand back or stay on

    You get a plain-language write-up. Or we stay on for hosting, upkeep and new work.

Often found together

Running Sentry alongside something else?

We work on the whole stack, so the problem can be wherever it actually is.

Questions

Sentry questions

Something else? Call+1 737 727 7327

Why is Sentry so noisy?

Almost always a handful of issues firing per visit with no fingerprint rule, non-error promise rejections, ResizeObserver warnings, aborted fetches from user navigation, or breadcrumbs from a chatty library. We write inbound filters and fingerprint rules so each real bug is one issue, not fifty, and the inbox becomes usable again.

Why are our stack traces unreadable?

Source maps are not being matched to the deployed bundle. The usual causes are a missing sentry-cli upload step in CI, a release name that does not match the one the SDK reports, or hashed filenames whose paths differ between build and upload. We wire the release and upload into CI and verify against a real event before calling it done.

How do we cut our Sentry bill without losing signal?

Three levers in order: inbound filters to drop noise before it is ingested, grouping fingerprints so one bug is one issue, and sampling on transactions and replays set to a rate that still gives you the picture you need. Most teams cut their events by half or more without missing a real regression once this is done properly.

Should we self-host Sentry or stay on SaaS?

Stay on SaaS unless data residency, regulatory or cost reasons genuinely tip it the other way. Self-hosted Sentry is around 20 services with Snuba, ClickHouse and Kafka, and the ops cost tends to exceed a mid-tier SaaS plan long before people admit it. We will do the arithmetic honestly for your situation.

Why do we not see a release on an issue?

Either the release is not being created in Sentry during CI, or the SDK is reporting a release string that does not match. We check the Sentry release API output and the SDK init call side by side, and usually find a small mismatch in how the Git SHA or version string is constructed.

How much does Sentry help cost?

The initial look is free. Share the Sentry org and the project that is misbehaving, and we return a scope with a number before any changes are made to the configuration.

Free estimate

Sentry giving you grief?

Tell us what you're running and what's going wrong. A person reads it and comes back with an estimate.

  • No charge to find out what it will take
  • If it's urgent, say so

Replies come from a person. By sending, you authorize us to email you back. No marketing.