# Troubleshooting

## No SSH keys found

<img width="492" alt="SSH key error" src="https://github.com/user-attachments/assets/3f7f9983-f92d-4900-9321-9d46c6355c36" />

Check your local ssh agent has your ssh key by running:

```shell
ssh-add -L
```

If your key isn't present, run the following on your local machine:

```shell
ssh-add
```

## Unable to find Python modules

```shell
rm -rf .venv
uv sync
```

Then {ref}`follow the steps <back-end>` if problems persist.

## Missing distutils package

Remove the virtual environment and resync dependencies:

```shell
rm -rf .venv
uv sync
```


## Event synchronization is not supported during trace capture

Report generation in TTNN is not compatible with trace capture. If you see this error,
it likely means that trace capture has been enabled in TTNN. This is not
a setting in TT-Metal or TTNN, but rather a hard-coded value in several of the models
and demos.

How trace capture is enabled depends on the model or demo. In some cases you will see
`enable_trace = True` in the Python code. In other cases it may be enabled by a
`--enable_trace` CLI arg. Some tests use the PyTest `parametrize` decorator to run the
demo both with and without trace capture.

Models or demos that have trace capture enabled need to be modified to not use that
feature in order for report generation to work.


## Pytest Timeout

Report generation when running models with `pytest` can sometimes take longer than the
default timeout value:

```
Failed: Timeout (>300.0s) from pytest-timeout.
```

This can be resolved by passing a `--timeout <timeout>` value to the `pytest` command,
with a larger timeout value than the default. For example:

```shell
python -m tracy -r -v -m pytest --timeout 900 /path/to/demo.py
```

(cluster-descriptor-missing)=
## Missing cluster descriptor 

The cluster layout view reads `cluster_descriptor.yaml` from the same folder as `db.sqlite` for the active memory report (for example `${TT_METAL_HOME}/generated/ttnn/reports/<report_folder>/cluster_descriptor.yaml`), but sometimes it has not been generated.

Run topology to generate the cluster descriptor.

   ```shell
   "${TT_METAL_HOME}/build/tools/umd/topology"
   ```

By default it will be output into a temporary directory `{temp_directory}/umd_XXXXXX/` which will be picked up when you generate your memory report.