Local Store

The Xping SDK records every test run to a local store on disk. This happens in both local-only and Cloud mode, so the CLI works whether or not you have an account.

Location

The store is resolved in this order:

The XPING_LOCAL_STORE environment variable, if set.
A .xping folder at your repository root — the nearest ancestor of the test assembly containing .git, *.sln, or *.slnx.
A per-repository folder under your local application data, when no repository root is found or the root is not writable.

To see where it resolved to:

dotnet xping where

The repository root is preferred because flakiness history is only meaningful per repository. A single shared folder would blend unrelated projects and break running the CLI inside a repo and getting that repo's answer.

The walk starts from the test assembly's location rather than the current directory, because the working directory during dotnet test varies between the CLI, IDE runners, and CI agents.

Layout

<repo-root>/.xping/
├── .gitignore          # contains `*`
├── sessions/
│   ├── session-638912345678901234-a3f5e9c2.json.gz
│   └── …
└── state.json

Each run is one gzipped JSON document holding the whole session. One file per run means concurrent test projects never contend for a lock, and a run interrupted by a killed test host leaves no file at all rather than a partial one.

Git

The store hides itself. .xping/.gitignore contains a single *, which git honours for the whole directory. Nothing appears in git status, and your repository's own .gitignore is never modified — adding an entry there would show up as an unexplained diff in someone's next commit.

You do not need to do anything. If you prefer an explicit entry, adding .xping/ to your own .gitignore is harmless.

Retention

Applied after every write, oldest run first, until all three limits hold:

Limit	Default
Maximum runs	50
Maximum total size	50 MB
Maximum age	30 days

A 2,000-test suite stores roughly 170 KB per run, so 50 runs is about 8.5 MB. A 200-test suite is well under 1 MB.

To delete history manually, use xping clear.

Performance

The store is written once per run, during finalization, after the last test has finished. Nothing is added to the per-test path.

Measured on a 2,000-test suite:

Operation	Cost	When
Assembling the session	1.4 ms	Once per run
Writing the run	9.4 ms	Once per run
Per-test overhead	none	—

Reading and analysing history is paid by the CLI, not by your test run.

Environment variables

`XPING_LOCAL_STORE`

Overrides the store location.

export XPING_LOCAL_STORE=/var/tmp/xping-store

Both the SDK and the CLI read this variable, and they must agree: if you set it for your test run, set it for xping report too, or the CLI will look somewhere else and find nothing.

Useful for read-only checkouts, containers where the repository root is not writable, and keeping history outside a workspace that gets wiped between builds.

`XPING_NO_BANNER`

Suppresses the one-line retry-flake hint the SDK prints after a test run, and the cloud invitation in the CLI report. Set it to any non-empty value.

export XPING_NO_BANNER=1

Failure behaviour

The local store never fails your test run. A read-only checkout, a full disk, or a locked file degrades to "no local history" — the failure is logged at debug level and nothing else happens.

Corrupt or unreadable run files are skipped when reading, not treated as fatal. A file written by a newer SDK than the one reading it is also skipped, so upgrading and downgrading costs some history rather than breaking the report.

What is stored

A deliberately reduced projection of each test execution:

Test fingerprint and display name
Outcome and duration
Retry attempt, and whether the test passed on a retry
A short hash of the error message, for grouping recurring failures

Plus per-run metadata: session id, timestamps, environment name, test assembly, branch, commit SHA, and whether the run was in CI.

Stack traces, error message text, exception types, and source locations are not stored. The store holds what local analysis needs and nothing more, which keeps it roughly ten times smaller than the uploaded payload.

A consequence worth knowing: because the local record is reduced, it cannot be re-uploaded as a full session. It is a source of local analysis, not a backup of your uploaded data.

Table of Contents