Security in Error Tracking
Who may do what with projects, issues and events (IAM), which credentials exist and what each may do, and what is recorded in the audit log.
IAM actions
Every request through the API, the CLI, Terraform or the console is checked against your IAM
policies. A project's resource name is rn:ews:de1:<organization id>:errors:project/<project id>.
Issues, events, transactions, profiles, queries, release health, alert rules and connections are
all checked against their project.
| Action | Access | What it allows |
|---|---|---|
errors:ListProjects | Read | List Error Tracking projects with their DSNs and event counts |
errors:CreateProject | Write | Create a project and its first DSN key |
errors:UpdateProject | Write | Change a project: name, platform, retention, rate limit, client IP, alert settings, DSN keys |
errors:DeleteProject | Write · destructive | Delete a project with its keys, issues, events and source maps |
errors:ListIssues | Read | List and read issues of a project: counts, users, first and last seen |
errors:UpdateIssue | Write | Resolve, ignore, reopen or assign issues |
errors:DeleteIssue | Write · destructive | Delete an issue and its events |
errors:GetEvents | Read | Read the events of an issue: stack traces, breadcrumbs, request, tags, contexts |
errors:UploadSourceMaps | Write | Create releases and upload source maps and artifact bundles (sentry-cli) |
errors:ListReleases | Read | List releases, their deploys and uploaded files |
errors:ListAlertRules | Read | List a project’s alert rules and what they fired |
errors:PutAlertRule | Write | Create or change an alert rule (trigger, filters, mail and webhook actions), or send its test alert |
errors:DeleteAlertRule | Write · destructive | Delete an alert rule and its history |
Generated from the service’s IAM descriptor. Sensitive reads hand over secret material or source code and are recorded in the audit log like writes; destructive writes delete, overwrite or revoke.
These are checked against the service, rn:ews:de1:<organization id>:errors:*, not a project:
errors:ListProjectsto list projects (the list then shows only the projects you may list) anderrors:CreateProjectto create one;errors:ListReleasesto list releases;- for an auth token,
errors:UploadSourceMapsandiam:CreateAccessKey; - through the API and the CLI,
errors:GetEventsto read a trace, a transaction's profile, or one statement across projects; the answer then holds only the projects whose events you may read.
Connecting a CI environment also needs ci:PutEnvironment on the repository; connecting an app,
deploy:ManageErrorsBinding on the app; connecting a database, rds:DescribeDBQueryInsights on
the database. A connected database grants nothing: each reader's own
rds:DescribeDBQueryInsights on it decides whether its statements are shown (see
Connect a database to a project).
A policy for someone who may read and triage one project's issues (replace org-id and project-id):
{
"Version": "2024-01-01",
"Statement": [
{
"Effect": "Allow",
"Action": ["errors:ListProjects", "errors:ListIssues", "errors:GetEvents", "errors:UpdateIssue"],
"Resource": "rn:ews:de1:org-id:errors:project/project-id"
},
{
"Effect": "Allow",
"Action": ["errors:ListProjects"],
"Resource": "rn:ews:de1:org-id:errors:*"
}
]
}Credentials
| Credential | May | Lives |
|---|---|---|
| DSN key | Send events to its project. Never read anything; safe in a browser bundle | Until revoked |
| Auth token | Everything its owner may do in Error Tracking (errors:*), nothing else |
The access-key lifetime it was given |
| Upload token | Upload source maps to one project, and the reads sentry-cli needs there |
At most an hour; a build's token 30 minutes, revoked when the build ends; a CI run's revoked when the run ends |
A connected project has a service account, errors-uploader-<slug>-…, whose policy allows only
the upload to that project. It holds no lasting key: every build and CI job gets a key of its own.
Deleting the project removes it.
Audit
Every change (creating, changing and deleting projects, keys, alert rules and connections,
triaging and deleting issues, rotating keys, creating tokens) is recorded in the audit log,
allowed or denied, once the request has passed validation. Most changes to a project look the
project up first: for a project id that does not exist (or is not a UUID) they answer 404 and
record nothing. Connecting a CI environment checks ci:PutEnvironment on the repository before it
looks for the project, so that decision is recorded either way. A CI job's token exchange is
recorded as an anonymous errors:UploadSourceMaps with the repository, environment and run.
Reads are recorded only when they are denied, at most once a minute per user and action.
Missing or not yours
Reading a project, its usage, or one issue answers 404 both when it does not exist and when you
may not read it. Most other reads of a project (issues, events, transactions, alert rules) answer 404 when
it does not exist and 403 when you may not read it.
Network exposure
The ingest at https://errors.<domain> accepts events from anywhere that holds a DSN key, browsers
included. The Sentry API on the same host answers only an authenticated request: a token, or an
access key in an Authorization: EWS-ACCESS-KEY header; it applies your IAM policies as the
Deutsche Cloud API does.