* fix(sandbox): add GPU device nodes and nvidia-persistenced to landlock baseline
Landlock READ_FILE/WRITE_FILE restricts open(2) on character device files
even when DAC permissions would otherwise allow it. GPU sandboxes need
/dev/nvidiactl, /dev/nvidia-uvm, /dev/nvidia-uvm-tools, /dev/nvidia-modeset,
and per-GPU /dev/nvidiaX nodes in the policy to allow NVML initialization.
Additionally, CDI bind-mounts /run/nvidia-persistenced/socket into the
container. NVML tries to connect to this socket at init time; if the
directory is not in the landlock policy, it receives EACCES (not
ECONNREFUSED), which causes NVML to abort with NVML_ERROR_INSUFFICIENT_PERMISSIONS
even though nvidia-persistenced is optional.
Both classes of paths are auto-added to the baseline when /dev/nvidiactl is
present. Per-GPU device nodes are enumerated at runtime to handle multi-GPU
configurations.
Replace manual Header::set_path() + append() with
builder.append_path_with_name() which emits GNU LongName extensions
for paths exceeding the 100-byte POSIX tar name field limit.
Fixes#705
Co-authored-by: John Myers <johntmyers@users.noreply.github.com>
The image push pipeline buffered the entire Docker image tar 3x in
memory (export, tar wrap, Bytes copy), causing OOM kills for images
over ~1-2 GB. Replace the in-memory pipeline with a temp-file +
streaming upload: export to a NamedTempFile, then stream the outer
tar (header, 8 MiB file chunks, footer) directly into
upload_to_container via body_try_stream. Peak memory drops from ~3x
image size to ~8 MiB constant.
Also adds incremental export progress reporting every 100 MiB.
* feat(bootstrap): switch GPU injection to CDI where supported
Use an explicit CDI device request (driver="cdi", device_ids=["nvidia.com/gpu=all"])
when the Docker daemon reports CDI spec directories via GET /info (SystemInfo.CDISpecDirs).
This makes device injection declarative and decouples spec generation from consumption.
When the daemon reports no CDI spec directories, fall back to the legacy NVIDIA device
request (driver="nvidia", count=-1) which relies on the NVIDIA Container Runtime hook.
Failure modes for both paths are equivalent: a missing or stale NVIDIA Container Toolkit
installation will cause container start to fail.
CDI spec generation is out of scope for this change; specs are expected to be
pre-generated out-of-band, for example by the NVIDIA Container Toolkit.
---------
Signed-off-by: Evan Lezar <elezar@nvidia.com>
Co-authored-by: Piotr Mlocek <pmlocek@nvidia.com>
* fix(cli): clear stale last-used sandbox on deletion
When a sandbox is deleted, the locally stored last-used record now gets
cleared if it matches the deleted sandbox name. This prevents subsequent
commands from falling back to a sandbox that no longer exists, which
previously caused confusing gRPC errors.
Adds clear_last_sandbox_if_matches() to openshell-bootstrap and calls
it from sandbox_delete() after each successful deletion.
Closes#172
Signed-off-by: Serge Panev <spanev@nvidia.com>
* style: fix rustfmt import ordering and line wrapping
Signed-off-by: Serge Panev <spanev@nvidia.com>
* fix(cli): pass gateway to sandbox_delete in finalize_sandbox_create_session
Signed-off-by: John Myers <johntmyers@users.noreply.github.com>
---------
Signed-off-by: Serge Panev <spanev@nvidia.com>
Signed-off-by: John Myers <johntmyers@users.noreply.github.com>
Co-authored-by: John Myers <johntmyers@users.noreply.github.com>
Error messages and recovery hints were showing 'openshell gateway destroy <name>'
as a positional argument, but the CLI defines name as a --name flag. This caused
'unexpected argument' errors when users copied the suggested commands.
Docker Desktop 29.x defaults to private cgroupns which prevents k3s
kubelet from accessing cgroup v2 controllers (cpu, cpuset, memory,
pids, hugetlb). This causes ContainerManager to fail during startup.
Explicitly set cgroupns_mode to host, which is backwards compatible
with all Docker versions and matches what k3s-in-Docker tooling
(k3d) requires.
Closes#273
Verify inference endpoints synchronously on the server during set/update, expose a --no-verify escape hatch in the CLI and Python helper, and return actionable failures when validation does not pass.
PR #281 removed the shared openshell-cluster Docker network in favor of
the default bridge. This restores custom bridge networking but makes each
gateway use its own isolated network named openshell-cluster-{name},
matching the existing container/volume naming convention.
Changes:
- Add network_name() to constants.rs for per-gateway network naming
- Add ensure_network() with retry/backoff and force_remove_network()
parameterized by network name instead of a global constant
- Attach containers to their per-gateway network via network_mode
- Disconnect and remove the network during gateway destroy
- Wire ensure_network() into the deploy flow before ensure_volume()
- Update architecture docs to reflect per-gateway network isolation
Drew Newberry