The docker module used different code for socket-activated docker daemon than for the non-socket activated daemon.
In particular, if the socket-activated daemon is used, then modprobe wasn't set up to be usable and in PATH for
the docker daemon, which resulted in a failure to start the daemon with overlayfs as storageDriver if the
`overlay` kernel module wasn't already loaded. This commit fixes that bug (which only appears if socket
activation is used), and also reduces the duplication between code paths so that it's easier to keep
both in sync in future.
As @domenkozar noted in #10828, cache=writeback seems to do more harm
than good:
https://github.com/NixOS/nixpkgs/issues/10828#issuecomment-164426821
He has tested it using the openstack NixOS tests and found that
cache=none significantly improves startup performance.
Signed-off-by: aszlig <aszlig@redmoonstudios.org>
This seems to be the root cause of the random page allocation failures
and @wizeman did a very good job on not only finding the root problem
but also giving a detailed explanation of it in #10828.
Here is an excerpt:
The problem here is that the kernel is trying to allocate a contiguous
section of 2^7=128 pages, which is 512 KB. This is way too much:
kernel pages tend to get fragmented over time and kernel developers
often go to great lengths to try allocating at most only 1 contiguous
page at a time whenever they can.
From the error message, it looks like the culprit is unionfs, but this
is misleading: unionfs is the name of the userspace process that was
running when the system ran out of memory, but it wasn't unionfs who
was allocating the memory: it was the kernel; specifically it was the
v9fs_dir_readdir_dotl() function, which is the code for handling the
readdir() function in the 9p filesystem (the filesystem that is used
to share a directory structure between a qemu host and its VM).
If you look at the code, here's what it's doing at the moment it tries
to allocate memory:
buflen = fid->clnt->msize - P9_IOHDRSZ;
rdir = v9fs_alloc_rdir_buf(file, buflen);
If you look into v9fs_alloc_rdir_buf(), you will see that it will try
to allocate a contiguous buffer of memory (using kzalloc(), which is a
wrapper around kmalloc()) of size buflen + 8 bytes or so.
So in reality, this code actually allocates a buffer of size
proportional to fid->clnt->msize. What is this msize? If you follow
the definition of the structures, you will see that it's the
negotiated buffer transfer size between 9p client and 9p server. On
the client side, it can be controlled with the msize mount option.
What this all means is that, the reason for running out of memory is
that the code (which we can't easily change) tries to allocate a
contiguous buffer of size more or less equal to "negotiated 9p
protocol buffer size", which seems to be way too big (in our NixOS
tests, at least).
After that initial finding, @lethalman tested the gnome3 gdm test
without setting the msize parameter at all and it seems to have resolved
the problem.
The reason why I'm committing this without testing against all of the
NixOS VM test is basically that I think we can only go better but not
worse than the current state.
Signed-off-by: aszlig <aszlig@redmoonstudios.org>
The most complex problems were from dealing with switches reverted in
the meantime (gcc5, gmp6, ncurses6).
It's likely that darwin is (still) broken nontrivially.
Commit 9bfe92ecee ("docker: Minor improvements, fix failing test") added
the services.docker.storageDriver option, made it mandatory but didn't
give it a default value. This results in an ugly traceback when users
enable docker, if they don't pay enough attention to also set the
storageDriver option. (An attempt was made to add an assertion, but it
didn't work, possibly because of how "mkMerge" works.)
The arguments against a default value were that the optimal value
depends on the filesystem on the host. This is, AFAICT, only in part
true. (It seems some backends are filesystem agnostic.) Also, docker
itself uses a default storage driver, "devicemapper", when no
--storage-driver=x options are given. Hence, we use the same value as
default.
Add a FIXME comment that 'devicemapper' breaks NixOS VM tests (for yet
unknown reasons), so we still run those with the 'overlay' driver.
Closes#10100 and #10217.
When using the ZFS storagedriver in docker, it shells out for the ZFS
commands. The path configuration for the systemd task does not include
ZFS, so if the driver is set to ZFS, add ZFS utilities to the PATH.
This will resolve https://github.com/NixOS/nixpkgs/issues/10127
[Bjørn: prefix commit message with "nixos/docker:", remove extra space
before ';']
The EBS and S3 (instance-store) AMIs are now created from the same
image. HVM instance-store AMIs are also generated.
Disk image generation has been factored out into a function
(nixos/lib/make-disk-image.nix) that can be used to build other kinds
of images.
Booting the demo/installer image won't work if the video memory is too
low. It boots into KDE, shows the background image and doesn't do
anything, according to @domenkozar.
Thanks to @domenkozar for reporting and testing this with 32MB.
Signed-off-by: aszlig <aszlig@redmoonstudios.org>
pkgs/os-specific/linux/kernel/common-config.nix defines HIGHMEM64G on
line 441 for 32bit systems, which implies PAE.
We now creating the OVA with PAE support enabled, which fixes bootup of
the image if people are just importing it without setting PAE
explicitly.
Signed-off-by: aszlig <aszlig@redmoonstudios.org>
Commit 687caeb renamed services.virtualboxHost to programs.virtualbox,
but according to the discussion on the commit, it's probably a better to
put it into virtualisation.virtualbox instead.
The discussion can be found here:
https://github.com/NixOS/nixpkgs/commit/687caeb#commitcomment-12664978
Signed-off-by: aszlig <aszlig@redmoonstudios.org>
This removes all references to .../sbin for the guest additions and also
installs all binaries to .../bin instead (so no more .../sbin).
The main motivation for doing this is commit 98cedb3 (which
unfortunately had to be reverted in a9f2e10) and pull request #9063,
where the latter is an initial effort to move mount.vboxsf to .../bin
instead of .../sbin.
The commit I made afterwards is finishing the removal of .../sbin
entirely.
In 14f09e0, I've introduced the module under modules/programs, because
the legacy virtualbox.nix was also under that path. But because we
already have modules/virtualisation/virtualbox-guest.nix, it really
makes sense to put this module alongside of it as well.
This module thus has no change in functionality and I've tested
evaluation against nixos/tests/virtualbox.nix and the manual.
Signed-off-by: aszlig <aszlig@redmoonstudios.org>
Using $storepath/sbin is deprecated according to commit 98cedb3, so
let's avoid putting anything in .../sbin for the guest additions.
This is a continuation of the initial commit done by @ctheune at
1fb1360, which unfortunately broke VM tests and only changed the path of
the mount.vboxsf helper.
With this commit, the VM test is fixed and I've also verified on my
machine that it is indeed working again.
Signed-off-by: aszlig <aszlig@redmoonstudios.org>
Xen required a few changes in order to be usable:
* Include xenfs module in initrd as loading it in the activation
script was failing.
* Include /etc/default/xendomains, which is needed by
xen-domains service.
* Create /var/log/xen and /var/lib/xen directories in
the xen-store service, which are needed by the xl command.
The directories could be created by any other script as long as
they are guaranteed to exist before xl is called.
* Fix a reference to /bin/ls in the xendomains script.
We already have separate tests for checking whether the ISO boots
correctly, so it's not necessary to do that here. So now
tests/installer.nix just tests nixos-install, from a regular NixOS VM
that uses the host's Nix store. This makes running the tests more
convenient because we don't have to build a new ISO after every
change.
The issue was that grub was not building the default entry which would
leave systems unbootable. This can now be safely reverted as the default
entry is being built once again.
This reverts commit fd1fb0403c.
If the host is shutting down, machinectl may fail because it's
bus-activated and D-Bus will be shutting down. So just send a signal
to the leader process directly.
Fixes#6212.
The Nixos Qemu VM that are used for VM tests can now start without
boot menu even when using a bootloader.
The Nixos Qemu VM with bootloader can emulate a EFI boot now.
- Create container nixos profile
- Create lxc-container nixos config using container nixos profile
- Docker nixos image, use nixos profile for its base config
Especially new users could be confused by this, so we're now marking
services.virtualbox.enable as obsolete and defaulting to
services.virtualboxGuest.enable instead. I believe this now makes it
clear, that this option is for guest additions only.
Signed-off-by: aszlig <aszlig@redmoonstudios.org>
Regression introduced in f496c3cbe4.
Previously when we used security.initialRootPassword, the default
priority for this option was 1001, because it was a default value set by
the option itself.
With the mentioned commit, it is no longer an option default but a
mkDefault, which is priority 1000.
I'm setting this to 150 now, as test-instrumentation.nix is using this
for overriding other options and because I think it still makes it
possible to simple-override it, because if no priority is given, we get
priority 100.
Signed-off-by: aszlig <aszlig@redmoonstudios.org>
This fixes the issue when the LXC emulator binary is garbage collected
and breaks libvirtd containers, because libvirtd XML file still refers
to GC'ed store path.
We already have a fix for QEMU, this commit extends the fix to cover LXC
too.
This tells the sad tale of @the-kenny who had bind-mounted his home
directory into a container. After doing `nixos-container destroy` he
discovered that his home directory went from "full of precious data" to
"no more data".
We want to avoid having similar sad tales in the future, so this now also
check this in the containers VM test.
Signed-off-by: aszlig <aszlig@redmoonstudios.org>
Fixes a leftover from 330fadb706.
We're using systemd dbus notifications now and this leftover caused the
startup notification to fail.
Signed-off-by: aszlig <aszlig@redmoonstudios.org>
This allows creating a container from an existing system store path,
which is especially nice for NixOps-deployed hosts because they don't
need a Nixpkgs tree anymore.
Systemd in a container will call sd_notify when it has finished
booting, so we can use that to signal that the container is
ready. This does require some fiddling with $NOTIFY_SOCKET.
Previously "machinectl reboot/poweroff" brutally killed the container,
as did "systemctl stop/restart". And reboot didn't actually work. Now
everything is fine.
Previously "machinectl reboot/poweroff" brutally killed the container,
as did "systemctl stop/restart". And reboot didn't actually work. Now
everything is fine.
curl does not retry if it is unable to connect to the metadata server.
For some reason, when creating a new AMI with a recent nixpkgs, the
metadata server would not be available when fetch-ec2-data ran. Switching
to wget that can retry even on TCP connection errors solved this problem.
I also made the fetch-ec2-data depend on ip-up.target, to get it to start
a bit later.
Removed the 'wait for GCE metadata service' job, as it was causing
issues with the metadata service (likely some firewall or something).
In stead, use wget with retries (including connection refused) in
stead or curl for fetching the SSH keys. Also made the stdout/-err
of this job appear in the console.
This version of module has disabled socketActivation, because until
nixos upgrade systemd to at least 214, systemd does not support
SocketGroup. So socket is created with "root" group when
socketActivation enabled. Should be fixed as soon as systemd upgraded.
Includes changes from #3015 and supersedes #3028
By setting a line like
MACVLANS="eno1"
in /etc/containers/<name>.conf, the container will get an Ethernet
interface named mv-eno1, which represents an additional MAC address on
the physical eno1 interface. Thus the container has direct access to
the physical network. You can specify multiple interfaces in MACVLANS.
Unfortunately, you can't do this with wireless interfaces.
Note that dhcpcd is disabled in containers by default, so you'll
probably want to set
networking.useDHCP = true;
in the container, or configure a static IP address.
To do: add a containers.* option for this, and a flag for
"nixos-container create".
Fixes#2379.
The new name was a misnomer because the values really are X11 video
drivers (e.g. ‘cirrus’ or ‘nvidia’), not OpenGL implementations. That
it's also used to set an OpenGL implementation for kmscon is just
confusing overloading.
By default, socat only waits 0.5s for the remote side to finish after
getting EOF on the local side. So don't close the local side, instead
wait for socat to exit when the remote side finishes.
http://hydra.nixos.org/build/10663282
By enabling ‘services.openssh.startWhenNeeded’, sshd is started
on-demand by systemd using socket activation. This is particularly
useful if you have a zillion containers and don't want to have sshd
running permanently. Note that socket activation is not noticeable
slower, contrary to what the manpage for ‘sshd -i’ says, so we might
want to make this the default one day.
The ability for unprivileged users to mount external media is useful
regardless of the desktop environment. Also, since udisks2 is
activated on-demand, it doesn't add any overhead if you're not using it.
Apparently systemd is now smart enough to figure out predictable names
for QEMU network interfaces. But since our tests expect them to be
named eth0/eth1..., this is not desirable at the moment.
http://hydra.nixos.org/build/10418789
