SHA256
1
0
forked from pool/apptainer
apptainer/apptainer.changes

257 lines
16 KiB
Plaintext

-------------------------------------------------------------------
Thu Aug 4 12:31:33 UTC 2022 - Christian Goll <cgoll@suse.com>
- Updated to version 1.1.0-rc1 which enables apptainer to run without
suid and additional groups. Although this is a prerelease this is
a major advantage justifying its use.
* Added a squashfuse image driver that enables mounting SIF files without
using setuid-root. Requires the squashfuse command and unprivileged user
namespaces.
* Added a fuse2fs image driver that enables mounting EXT3 files and EXT3 SIF
overlay partitions without using setuid-root. Requires the fuse2fs command
and unprivileged user namespaces.
* Added the ability to use persistent overlay (--overlay) and
--writable-tmpfs without using setuid-root. This requires unprivileged user
namespaces and either a new enough kernel (>= 5.11) or the fuse-overlayfs
command. Persistent overlay works when the overlay path points to a regular
filesystem (known as "sandbox" mode, which is not allowed when in setuid
mode), or when it points to an EXT3 image. Does not work with a SIF
partition because that requires privileges to mount as an ext3 image.
* Extended the --fakeroot option to be useful when /etc/subuid and
/etc/subgid mappings have not been set up. If they have not been set up, a
root-mapped unprivileged user namespace (the equivalent of unshare -r)
and/or the fakeroot command from the host will be tried. Together they
emulate the mappings pretty well but they are simpler to administer. This
feature is especially useful with the --overlay and --writable-tmpfs
options and for building containers unprivileged, because they allow
installing packages that assume they're running as root. A limitation on
using it with --overlay and --writable-tmpfs however is that when only the
fakeroot command can be used (because there are no user namespaces
available, in suid mode) then the base image has to be a sandbox. This
feature works nested inside of an apptainer container, where another
apptainer command will also be in the fakeroot environment without
requesting the --fakeroot option again, or it can be used inside an
apptainer container that was not started with --fakeroot. However, the
fakeroot command uses LD_PRELOAD and so needs to be bound into the
container which requires a compatible libc. For that reason it doesn't work
when the host and container operating systems are of very different
vintages. If that's a problem and you want to use only an unprivileged
root-mapped namespace even when the fakeroot command is installed, just run
apptainer with unshare -r.
* Made the --fakeroot option be implied when an unprivileged user builds a
container from a definition file. When /etc/subuid and /etc/subgid mappings
are not available, all scriptlets are run in a root-mapped unprivileged
namespace (when possible) and the %post scriptlet is additionally run with
the fakeroot command. When unprivileged user namespaces are not available,
such that only the fakeroot command can be used, the --fix-perms option is
implied to allow writing into directories.
* Added a --fakeroot option to the apptainer overlay create command to make
an overlay EXT3 image file that works with the fakeroot that comes from
unprivileged root-mapped namespaces. This is not needed with the fakeroot
that comes with /etc/sub[ug]id mappings nor with the fakeroot that comes
with only the fakeroot command in suid flow.
* $HOME is now used to find the user's configuration and cache by default. If
that is not set it will fall back to the previous behavior of looking up
the home directory in the password file. The value of $HOME inside the
container still defaults to the home directory in the password file and can
still be overridden by the --home option.
* When starting a container, if the user has specified the cwd by using the
--pwd flag, if there is a problem an error is returned instead of
defaulting to a different directory.
* Nesting of bind mounts now works even when a --bind option specified a
different source and destination with a colon between them. Now the
APPTAINER_BIND environment variable makes sure the bind source is from the
bind destination so it will be succesfully re-bound into a nested apptainer
container.
* The warning about more than 50 bind mounts required for an underlay bind
has been changed to an info message.
* oci mount sets Process.Terminal: true when creating an OCI config.json, so
that oci run provides expected interactive behavior by default.
The default hostname for oci mount containers is now apptainer instead of mrsdalloway.
* systemd is now supported and used as the default cgroups manager. Set
systemd cgroups = no in apptainer.conf to manage cgroups directly via the
cgroupfs.
* Added a new action flag --no-eval which:
+ Prevents shell evaluation of APPTAINERENV_ / --env / --env-file
environment variables as they are injected in the container, to match
OCI behavior. Applies to all containers.
+ Prevents shell evaluation of the values of CMD / ENTRYPOINT and command
line arguments for containers run or built directly from an OCI/Docker
source. Applies to newly built containers only, use apptainer inspect
to check version that container was built with.
* Added --no-eval to the list of flags set by the OCI/Docker --compat mode.
* sinit process has been renamed to appinit.
* Added --keysdir to key command to provide an alternative way of setting
local keyring path. The existing reading of the keyring path from
environment variable 'APPTAINER_KEYSDIR' is untouched.
* apptainer key push will output the key server's response if included in
order to help guide users through any identity verification the server may
require.
* ECL no longer requires verification for all signatures, but only when
signature verification would alter the expected behavior of the list:
+ At least one matching signature included in a whitelist must be
validated, but other unvalidated signatures do not cause ECL to fail.
+ All matching signatures included in a whitestrict must be validated,
but unvalidated signatures not in the whitestrict do not cause ECL to
fail.
+ Signature verification is not checked for a blacklist; unvalidated
signatures can still block execution via ECL, and unvalidated
signatures not in the blacklist do not cause ECL to fail.
- New features / functionalities
* Non-root users can now use --apply-cgroups with run/shell/exec to limit
container resource usage on a system using cgroups v2 and the systemd
cgroups manager.
* Native cgroups v2 resource limits can be specified using the [unified] key
in a cgroups toml file applied via --apply-cgroups.
* Added --cpu*, --blkio*, --memory*, --pids-limit flags to apply cgroups
resource limits to a container directly.
Added instance stats command.
* The --no-mount flag & APPTAINER_NO_MOUNT env var can now be used to disable
a bind path entry from apptainer.conf by specifying the absolute path to
the destination of the bind.
* Apptainer now supports the riscv64 architecture.
* remote add --insecure may now be used to configure endpoints that are only
accessible via http. Alternatively the environment variable
APPTAINER_ADD_INSECURE can be set to true to allow http remotes to be added
wihtout the --insecure flag. Specifying https in the remote URI overrules
both --insecure and APPTAINER_ADD_INSECURE.
* Gpu flags --nv and --rocm can now be used from an apptainer nested inside
another apptainer container.
* Added --public, --secret, and --both flags to the key remove command to
support removing secret keys from the apptainer keyring.
* Debug output can now be enabled by setting the APPTAINER_DEBUG env var.
* Debug output is now shown for nested apptainer calls, in wrapped unsquashfs
image extraction, and build stages.
- Bug fixes
* Remove warning message about SINGULARITY and APPTAINER variables having
different values when the SINGULARITY variable is not set.
* Add specific error for unreadable image / overlay file.
* Pass through a literal \n in host environment variables to the container.
* Fix loop device creation with loop-control when running inside docker containers.
* Fix the issue that the oras protocol would ignore the --no-https/--nohttps flag.
- File changes
* Removed useful_error_message.patch as not needed any more
* Added fix-32bit-compilation.patch from upstream
-------------------------------------------------------------------
Mon Jul 11 09:38:45 UTC 2022 - Christian Goll <cgoll@suse.com>
- Update to version 1.0.3:
* Process redirects that can come from sregistry with a library:// URL.
* Fix inspect --deffile and inspect --all to correctly show definition files
in sandbox container images instead of empty output. This has a side effect
of also fixing the storing of definition files in the metadata of sif files
built by Apptainer, because that metadata is constructed by doing inspect
--all.
-------------------------------------------------------------------
Wed May 18 12:07:59 UTC 2022 - Dominique Leuenberger <dimstar@opensuse.org>
- Update to version 1.0.2:
+ Fixed `FATAL` error thrown by user configuration migration code
that caused users with inaccessible home directories to be
unable to use `apptainer` commands.
+ Do not truncate environment variables with commas.
+ Use HEAD request when checking digest of remote OCI image
sources, with GET as a fall-back. Greatly reduces Apptainer's
impact on Docker Hub API limits.
-------------------------------------------------------------------
Fri Mar 18 16:02:59 UTC 2022 - Christian Goll <cgoll@suse.com>
- Updated to v1.0.1 with following bug fixes
* Don't prompt for y/n to overwrite an existing file when build is called
from a non-interactive environment. Fail with an error.
* Preload NSS libraries prior to mountspace name creation to avoid
circumstances that can cause loading those libraries from the container
image instead of the host, for example in the startup environment.
* Fix race condition where newly created loop devices can sometimes not be opened.
* Support nvidia-container-cli v1.8.0 and above, via fix to capability set.
-------------------------------------------------------------------
Thu Feb 17 15:29:45 UTC 2022 - Christian Goll <cgoll@suse.com>
- Updated to v1.0.0-rc1 changes to singularity 3.9.5 are
* The primary executable has been changed from singularity to apptainer.
However, a singularity command symlink alias has been created pointing to
the apptainer command. The contents of containers are unchanged and
continue to use the singularity name for startup scripts, etc.
* The per-user configuration directory has changed from ~/.singularity to
~/.apptainer. The first time the apptainer command accesses the user
configuration directory, relevant configuration is automatically imported
from the old directory to the new one.
* Environment variables have all been changed to have an APPTAINER prefix
instead of a SINGULARITY prefix. However, SINGULARITY prefix variables are
still recognized. If only a SINGULARITY prefix variable exists, a warning
will be printed about deprecated usage and then the value will be used. If
both prefixes exist and the value is the same, no warning is printed; this
is the recommended method to set environment variables for those who need
to support both apptainer and singularity. If both prefixes exist for the
same variable and the value is different then a warning is also printed.
* The default SylabsCloud remote endpoint has been removed and replaced by
one called DefaultRemote which has no defined server for the library://
URI. System administrators may restore the old default if they wish by
adding it to /etc/apptainer/remote.yaml with a URI of cloud.sylabs.io and
setting it there as the Active remote, or users can add it to their own
configuration with the commands apptainer remote add SylabsCloud
cloud.sylabs.io and apptainer remote use SylabsCloud.
* The DefaultRemote's key server is https://keys.openpgp.org instead of the
Sylabs key server
* The apptainer build --remote option has been removed because there is no
standard protocol or non-commercial service that supports it.
- New Features:
* Honor image binds and user binds in the order they're given instead of
always doing image binds first.
* Experimental support for checkpointing of instances using DMTCP has been
added. Additional flags --dmtcp-launch and --dmtcp-restart has been added
to the apptainer instance start command, and a checkpoint command group has
been added to manage the checkpoint state. A new
/etc/apptainer/dmtcp-conf.yaml configuration file is also added.
Limitations are that it can only work with dynamically linked applications
and the container has to be based on glibc.
* --writable-tmpfs can be used with apptainer build to run the %test section
of the build with a ephemeral tmpfs overlay, permitting tests that write to
the container filesystem.
* The --compat flag for actions is a new short-hand to enable a number of
options that increase OCI/Docker compatibility. Infers --containall,
--no-init, --no-umask, --writable-tmpfs. Does not use user, uts, or network
namespaces as these may not be supported on many installations.
* The experimental --nvccli flag will use nvidia-container-cli to setup the
container for Nvidia GPU operation. Apptainer will not bind GPU libraries
itself. Environment variables that are used with Nvidia's docker-nvidia
runtime to configure GPU visibility / driver capabilities & requirements
are parsed by the --nvccli flag from the environment of the calling user.
By default, the compute and utility GPU capabilities are configured. The
use nvidia-container-cli option in apptainer.conf can be set to yes to
always use nvidia-container-cli when supported. --nvccli is not supported
in the setuid workflow, and it requires being used in combination with
--writable in user namespace mode. Please see documentation for more
details.
* The --apply-cgroups flag can be used to apply cgroups resource and device
restrictions on a system using the v2 unified cgroups hierarchy. The
resource restrictions must still be specified in the v1 / OCI format, which
will be translated into v2 cgroups resource restrictions, and eBPF device
restrictions.
* A new --mount flag and APPTAINER_MOUNT environment variable can be used to
specify bind mounts in
type=bind,source=<src>,destination=<dst>[,options...] format. This improves
CLI compatibility with other runtimes, and allows binding paths containing
: and , characters (using CSV style escaping).
* Perform concurrent multi-part downloads for library:// URIs. Uses 3
concurrent downloads by default, and is configurable in apptainer.conf or
via environment variables.
-------------------------------------------------------------------
Wed Dec 15 08:38:57 UTC 2021 - Christian Goll <cgoll@suse.com>
- Explicit dependcy on go1.16.12 or go1.17.5 which fix
(CVE-2021-44717) and (CVE-2021-44716) that may affect singualrity
-------------------------------------------------------------------
Mon Dec 13 12:55:47 UTC 2021 - Christian Goll <cgoll@suse.com>
- inital commit of apptainer which is a singularity fork