A failure to reach the control node can cause slurmd to fail and the
unit remains in the failed state until is manually restarted. Instead,
try to restart the service every 30 seconds, forever:
owl1% systemctl show slurmd | grep -E 'Restart=|RestartUSec='
Restart=on-failure
RestartUSec=30s
owl1% pgrep slurmd
5903
owl1% sudo kill -SEGV 5903
owl1% pgrep slurmd
6137
Fixes: #177
Reviewed-by: Aleix Boné <abonerib@bsc.es>
We change the search procedure so it detects NixOS from /etc/os-release
and uses "libnuma.so" when calling dlopen, instead of harcoding a full
path to /usr. The full patch of libnuma is stored in the runpath, so
dlopen can find it.
Reviewed-by: Aleix Boné <abonerib@bsc.es>
Tested-by: Vincent Arcila <vincent.arcila@bsc.es>
It tries to dlopen libcrypt.so.1 and libstdc++.so.6, so we make sure
they are available by adding them to the runpath.
Reviewed-by: Aleix Boné <abonerib@bsc.es>
The hrtimer_init() is now done via hrtimer_setup() with the callback
function as argument.
See: https://lwn.net/Articles/996598/
Reviewed-by: Aleix Boné <abonerib@bsc.es>
Allows direct contact via the VPN when accessing from fox, but use
Internet when using the rest of the machines.
Reviewed-by: Aleix Roca Nonell <aleix.rocanonell@bsc.es>
The StartLimitBurst and StartLimitIntervalSec options belong to the
[Unit] section, otherwise they are ignored in [Service]:
> Unknown key 'StartLimitIntervalSec' in section [Service], ignoring.
When using [Unit], the limits are properly set:
apex% systemctl show power-policy.service | grep StartLimit
StartLimitIntervalUSec=10min
StartLimitBurst=10
StartLimitAction=none
Reviewed-by: Aleix Boné <abonerib@bsc.es>
In all machines, as soon as we recover the power, turn the machine back
on. We cannot rely on the previous state as we will shut them down
before the power is cut to prevent damage on the power supply
monitoring circuit.
Reviewed-by: Aleix Roca Nonell <aleix.rocanonell@bsc.es>
Reviewed-by: Aleix Boné <abonerib@bsc.es>
The UPC has different dates for the yearly power cut, and Fox can
recover properly from a power loss, so we don't need to have it turned
off before the power cut. Simply disabling the timer is enough.
Reviewed-by: Aleix Roca Nonell <aleix.rocanonell@bsc.es>
Reviewed-by: Aleix Boné <abonerib@bsc.es>
The cudainfo program checks that we can initialize the CUDA RT library
and communicate with the driver. It can be used as standalone program or
built with cudainfo.gpuCheck so it is executed inside the build sandbox
to see if it also works fine. It uses the autoAddDriverRunpath hook to
inject in the runpath the location of the library directory for CUDA
libraries.
Reviewed-by: Aleix Boné <abonerib@bsc.es>
This allows running derivations which depend on cuda runtime without
breaking the sandbox. We only need to add `requiredSystemFeatures = [ "cuda" ];`
to the derivation.
Reviewed-by: Rodrigo Arias Mallo <rodrigo.arias@bsc.es>
The NVIDIA GTX 960 from 2016 has the Maxwell architecture, and NixOS
suggests using the proprietary driver for older than Turing:
> It is suggested to use the open source kernel modules on Turing or
> later GPUs (RTX series, GTX 16xx), and the closed source modules
> otherwise.
Reviewed-by: Aleix Boné <abonerib@bsc.es>
Fails with: "error: ix has been removed from Nixpkgs, as the ix.io
pastebin has been offline since Dec. 2023".
Reviewed-by: Aleix Roca Nonell <aleix.rocanonell@bsc.es>
Access internal hosts via apex proxy. From the compute nodes we first
open an SSH connection to apex, and then tunnel it through the HTTP
proxy with netcat.
This way we allow reaching internal GitLab repositories without
requiring the user to have credentials in the remote host, while we can
use multiple remotes to provide redundancy.
Reviewed-by: Aleix Boné <abonerib@bsc.es>
Don't wait to flush writes, as we don't care about consistency on a
crash:
> This option allows the NFS server to violate the NFS protocol and
> reply to requests before any changes made by that request have been
> committed to stable storage (e.g. disc drive).
>
> Using this option usually improves performance, but at the cost that
> an unclean server restart (i.e. a crash) can cause data to be lost or
> corrupted.
Reviewed-by: Aleix Boné <abonerib@bsc.es>
Get rid of all the spam accounts they are trying to register.
Reviewed-by: Aleix Boné <abonerib@bsc.es>
Reviewed-by: Aleix Roca Nonell <aleix.rocanonell@bsc.es>
We moved the tent machine to the server room in the BSC building and is
now directly connected to the raccoon via NAT.
Fixes: #106
Reviewed-by: Aleix Boné <abonerib@bsc.es>
Allow xeon machines to optionally inherit SSF configuration such as the
NFS mount point and the network configuration.
Reviewed-by: Aleix Boné <abonerib@bsc.es>
From the nix manual:
> A configuration setting usually overrides any previous value. However,
> for settings that take a list of items, you can prefix the name of the
> setting by extra- to append to the previous value.
Reviewed-by: Rodrigo Arias Mallo <rodrigo.arias@bsc.es>
Documents the location, board and serial numbers so we can track the
machines if they move around. Some information is unkown.
Using the Nix language to encode the machines location and properties
allows us to later use that information in the configuration of the
machines themselves.
Reviewed-by: Aleix Boné <abonerib@bsc.es>
We will need to setup an VPN to be able to access fox in its new
location, so for now we simply remove the IPMI monitoring.
Reviewed-by: Aleix Boné <abonerib@bsc.es>
Fox should reply once the machine is connected to the UPC network.
Monitoring also the gateway and UPC anella allows us to estimate if the
whole network is down or just fox.
Reviewed-by: Aleix Boné <abonerib@bsc.es>
The fox machine will be placed in the UPC network, so we update the
configuration with the new IP and gateway. We won't be able to reach hut
directly so we also remove the host entry and proxy.
Reviewed-by: Aleix Boné <abonerib@bsc.es>
These sensors are part of their air quality measurements, which just
happen to be very close to our server room.
Reviewed-by: Aleix Boné <abonerib@bsc.es>
Allows us to track ambient temperature changes and estimate the
temperature delta between the server room and exterior temperature.
We should be able to predict when we would need to stop the machines due
to excesive temperature as summer approaches.
Reviewed-by: Aleix Boné <abonerib@bsc.es>
Allows us to see which derivations are being built in realtime. It is a
bit of a hack, but it seems to work. We simply look at the environment
of the child processes of nix-daemon (usually bash) and then look for
the $name variable which should hold the current derivation being
built. Needs root to be able to read the environ file of the different
nix-daemon processes as they are owned by the nixbld* users.
See: https://discourse.nixos.org/t/query-ongoing-builds/23486
Reviewed-by: Aleix Boné <abonerib@bsc.es>
From the documentation of keep-outputs, setting it to true would prevent
the GC from removing build time dependencies:
If true, the garbage collector will keep the outputs of non-garbage
derivations. If false (default), outputs will be deleted unless they are
GC roots themselves (or reachable from other roots).
In general, outputs must be registered as roots separately. However,
even if the output of a derivation is registered as a root, the
collector will still delete store paths that are used only at build time
(e.g., the C compiler, or source tarballs downloaded from the network).
To prevent it from doing so, set this option to true.
See: https://nix.dev/manual/nix/2.24/command-ref/conf-file.html#conf-keep-outputs
Reviewed-by: Aleix Roca Nonell <aleix.rocanonell@bsc.es>
Now that we have more space, we can extend the retention time to 5 years
to hold the monitoring metrics. For a year we have:
# du -sh /var/lib/prometheus2
13G /var/lib/prometheus2
So we can expect it to increase to about 65 GiB. In the future we may
want to reduce some adquisition frequency.
Reviewed-by: Aleix Boné <abonerib@bsc.es>
Allows fetching repositories from hut and other machines in jungle
without the need to do any extra configuration.
Reviewed-by: Aleix Boné <abonerib@bsc.es>
For owl1 and owl2 directly connect to hut via LAN with HTTP, but for
raccoon pass via the proxy using jungle.bsc.es with HTTPS. There is no
risk of tampering as packages are signed.
Reviewed-by: Rodrigo Arias Mallo <rodrigo.arias@bsc.es>
Prevents the "iptables: Chain already exists." error by making sure that
we don't leave any chain on start. The ideal solution is to use
iptables-restore instead, which will do the right job. But this needs to
be changed in NixOS entirely.
Reviewed-by: Aleix Boné <abonerib@bsc.es>
Needed for local hosts to contact the nix cache via HTTP directly.
We also allow the incoming traffic on port 80.
Reviewed-by: Aleix Boné <abonerib@bsc.es>
It uses docker based on alpine and the host nix store, so we can perform
builds but isolate them from the system.
Reviewed-by: Aleix Boné <abonerib@bsc.es>
Ensure that all hut users have a paste directory in /ceph/p owned by
themselves. We need to wait for the ceph mount point to create them, so
we use a systemd service that waits for the remote-fs.target.
Reviewed-by: Aleix Boné <abonerib@bsc.es>
Instead of using multiple tunels to forward all our services to the VM
that serves jungle.bsc.es, just use nginx to redirect the traffic from
hut. This allows adding custom rules for paths that are not posible
otherwise.
Reviewed-by: Aleix Boné <abonerib@bsc.es>
Prevents starting the garbage collector before the remote FS are
mounted, in particular /home. Otherwise, all the gcroots which have
symlinks in /home will be considered stale and they will be removed.
See: #79
Reviewed-by: Aleix Roca Nonell <aleix.rocanonell@bsc.es>
Apparently the ttyS0 console doesn't exist but ttyS1 does:
raccoon% sudo stty -F /dev/ttyS0
stty: /dev/ttyS0: Input/output error
raccoon% sudo stty -F /dev/ttyS1
speed 9600 baud; line = 0;
-brkint -imaxbel
The dmesg line agrees:
00:03: ttyS1 at I/O 0x2f8 (irq = 3, base_baud = 115200) is a 16550A
The console configuration is then moved from base to xeon to allow
changing it for the raccoon machine.
Reviewed-by: Aleix Boné <abonerib@bsc.es>
They have been removed from NixOS. The "hardware.opengl" group is now
renamed to "hardware.graphics".
See: 98cef4c273
Reviewed-by: Aleix Boné <abonerib@bsc.es>
The shutdown timer will fire at slightly different times for the
different nodes, so we slowly decrease the power consumption.
Reviewed-by: Aleix Boné <abonerib@bsc.es>
Initially we planned to run jobs in those nodes by sharing the same nix
store from hut. However, these nodes are now used to build packages
which are not available in hut. Users also ssh to the nodes, which
doesn't mount the hut store, so it doesn't make much sense to keep
mounting it.
Reviewed-by: Aleix Boné <abonerib@bsc.es>
The current select mechanism was using the memory too as a consumable
resource, which by default only sets 1 MiB per node. As each job already
requests 1 MiB, it prevents other jobs from running.
As we are not really concerned with memory usage, we only use the unused
cores in the select criteria.
Reviewed-by: Aleix Boné <abonerib@bsc.es>
Starting with GitLab 16, there is a new mechanism to authenticate the
runners via authentication tokens, so use it instead. Older tokens and
runners are also removed, as they are no longer used.
With the new way of managing tokens, both the tags and the locked state
are managed from the GitLab web page.
See: https://docs.gitlab.com/ee/ci/runners/new_creation_workflow.html
Reviewed-by: Aleix Boné <abonerib@bsc.es>
Allows users to attach GDB to their own processes, without requiring
running the program with GDB from the start. It is only available in
compute nodes, the storage nodes continue with the restricted settings.
Reviewed-by: Aleix Boné <abonerib@bsc.es>
smp_processor_id() was called in a preepmtible context, which could
invalidate the returned value. However, this was not harmful, because
fcs threads in nosv are pinned.
Reviewed-by: Rodrigo Arias Mallo <rodrigo.arias@bsc.es>
Access to other machines can be easily added into the "hosts" attribute
without the need to replicate the configuration.
Reviewed-by: Aleix Roca Nonell <aleix.rocanonell@bsc.es>
To accomodate the raccoon knights workstation, some of the configuration
pulled by m/common/main.nix has to be removed. To solve it, the xeon
specific parts are placed into m/common/xeon.nix and only the common
configuration is at m/common/base.nix.
Reviewed-by: Aleix Roca Nonell <aleix.rocanonell@bsc.es>
The users.jungleUsers configuration option behaves like the users.users
option, but defines the list attribute `hosts` for each user, which
filters users so that only the user can only access those hosts.
Reviewed-by: Aleix Roca Nonell <aleix.rocanonell@bsc.es>
The database will hold the performance results of the execution of the
benchmarks. We follow the same setup on knights3 for now.
Reviewed-by: Aleix Roca Nonell <aleix.rocanonell@bsc.es>
Allows sending Grafana alerts via email too, so we have a reduntant
mechanism in case Slack fails to deliver them.
Reviewed-by: Aleix Roca Nonell <aleix.rocanonell@bsc.es>
The firewall was blocking the monitoring traffic from hut and the Ceph
traffic among OSDs. The rules only allow connecting from the specific
host that they are supposed to be coming from.
Reviewed-by: Aleix Roca Nonell <aleix.rocanonell@bsc.es>
The main website of the BSC is failing every day around 3:00 AM for
almost one hour, so it is not a very good target. Instead, google.com is
used which should be more reliable. The same robots.txt path is fetched,
as it is smaller than the main page.
Reviewed-by: Aleix Roca Nonell <aleix.rocanonell@bsc.es>
As all other HTTPS probes pass through the opsproxy01.bsc.es proxy, we
cannot detect a problem in our proxy or in the BSC one. Adding another
target like bsc.es that doesn't use the ops proxy allows us to discern
where the problem lies.
Instead of monitoring https://www.bsc.es/ directly, which will trigger
the whole Drupal server and take a whole second, we just fetch robots.txt
so the overhead on the server is minimal (and returns in less than 10 ms).
Reviewed-by: Aleix Roca Nonell <aleix.rocanonell@bsc.es>
The public-inbox service fetches emails from the sourcehut mailing lists
and displays them on the web. The idea is to reduce the dependency on
external services and add a secondary storage for the mailing lists in
case sourcehut goes down or changes the current free plans.
The service is available in https://jungle.bsc.es/lists/ and is open to
the public. It currently mirrors the bscpkgs and jungle mailing list.
We also edited the CSS to improve the readability and have larger fonts
by default.
The service for public-inbox produced by NixOS is not well configured to
fetch emails from an IMAP mail server, so we also manually edit the
service file to enable the network.
Reviewed-by: Aleix Roca Nonell <aleix.rocanonell@bsc.es>
The GitLab instance is in the /gitlab endpoint and may fail
independently of https://pm.bsc.es/.
Cc: Víctor López <victor.lopez@bsc.es>
Reviewed-by: Aleix Roca Nonell <aleix.rocanonell@bsc.es>
The module is only enabled on Hut and Eudy because we noticed activity
on the debuginfod service even if no debug session was active.
Reviewed-by: Rodrigo Arias Mallo <rodrigo.arias@bsc.es>
The /tmp directory was using the SSD disk which is not erased across
boots. Nix will use /tmp to perform the builds, so we want it to be as
fast as possible. In general, all the machines have enough space to
handle large builds like LLVM.
Reviewed-by: Aleix Roca Nonell <aleix.rocanonell@bsc.es>
The old runners for the PM gitlab were disabled in configuration in the
last outage, but they remained working until we reboot the node. With
this change we enable the runners for both PM and gitlab.bsc.es.
Reviewed-by: Aleix Roca Nonell <aleix.rocanonell@bsc.es>
The target gw.bsc.es doesn't reply to our ICMP probes from hut. However,
the anella hop in the tracepath is a good candidate to identify cuts
between the login and the provider and between the provider and external
hosts like Google or Cloudflare DNS.
Reviewed-By: Aleix Roca Nonell <aleix.rocanonell@bsc.es>
These probes check if we can reach several targets via ICMP, which is
not proxied, so they can be used to see if ICMP forwarding is working in
the login node.
In particular, we test if we can reach the Google (8.8.8.8) and
Cloudflare (1.1.1.1) DNS servers, the BSC gateway which responds to ping
only from the intranet and the login node (ssfhead).
Reviewed-By: Aleix Roca Nonell <aleix.rocanonell@bsc.es>
The alerts need to contact the slack endpoint, so we add the proxy
environment variables to the grafana systemd service.
Reviewed-By: Aleix Roca Nonell <aleix.rocanonell@bsc.es>
By default it was trying to reach the targets using the default gateway,
but since the electrical cut of 2023-10-20, the login node has not
enabled forwarding again. So better if we don't rely on it.
Reviewed-By: Aleix Roca Nonell <aleix.rocanonell@bsc.es>
This is done to prevent accidental evaluations where the nixpkgs input
of bscpkgs is still pointing to a different version that the one
specified in the jungle flake. Instead use jungle#bscpkgs.X to get a
package from bscpkgs.
**What does this PR change? What problem does it solve?**
<!--
Describe the changes and their purpose here, as detailed as and if needed.
Please do not add 2 unrelated changes in a single PR as it is difficult to track/revert those in future.
-->
**Was the change discussed in an issue or in the Discussions before?**
<!--
Link issues and relevant Discussions posts here.
If this PR resolves an issue on GitHub, use "Closes #1234" so that the issue
is closed automatically when this PR is merged.
-->
## PR Checklist
- [ ] This change adds/updates translations and I have used the [template present here](https://github.com/adityatelange/hugo-PaperMod/wiki/Translations#want-to-add-your-language-).
- [ ] I have enabled [maintainer edits for this PR](https://help.github.com/en/github/collaborating-with-issues-and-pull-requests/allowing-changes-to-a-pull-request-branch-created-from-a-fork).
- [ ] I have verified that the code works as described/as intended.
- [ ] This change adds a Social Icon which has a permissive license to use it.
- [ ] This change **does not** include any CDN resources/links.
- [ ] This change **does not** include any unrelated scripts such as bash and python scripts.
- [ ] This change updates the overridden internal templates from HUGO's repository.
> Hugo PaperMod is a theme based on [hugo-paper](https://github.com/nanxiaobei/hugo-paper).
> The goal of this project is to add more features and customization to the og theme.
**Documentation** can be found here: [**📚 Wiki**](https://github.com/adityatelange/hugo-PaperMod/wiki)
**ExampleSite** can be found here: [**exampleSite**](https://github.com/adityatelange/hugo-PaperMod/tree/exampleSite). Demo is built up with [exampleSite](https://github.com/adityatelange/hugo-PaperMod/tree/exampleSite) as source.
[](https://github.com/gohugoio/hugo/releases/tag/v0.97.1)
Blocking a user prevents them from interacting with repositories, such as opening or commenting on pull requests or issues. Learn more about blocking a user.
