Starting in v3, more info here: https://github.com/immich-app/immich/discussions/29439

Love immich, but recently have been cleaning up my server and watching for stray processor and io use. This was one thing that jumped out. Personally, I don't want my full image library checked each night and fixed it to something that made sense for me.

Because my last post about just having some beers and discussing homelabbing with dutch people from this community got more attention than expected, we have actually set up an event!

First edition of homelabbing + beers NL will be held in Utrecht on the 21st of November.
Price will be 19 euros and drinks are included. This is to prevent that i have to pay a massive bill by the end lol. More info on exact location and time can be found on discord.

If you intend to show up, please set yourself as interested for the event in our newly created discord server: https://discord.gg/qbCrzfpMU
Looking forward to seeing you guys at the meetup

My wife and I have, for many years, used and relied on Google family location sharing in Google maps, to see each other's location. it worked well for many years, while we all had Android phones.

Recently, my wife has moved to an iPhone, and it no longer works. iOS is just a **** when it comes to arbitrarily putting apps to sleep. Despite giving Google maps every single permission and set it on the background always on for location etc, it still doesn't work properly.

Sometimes, after a few days of not using, I'll try to check if she made it safely to work, and it will say last reported location like 8 hours ago.

We then tried using Home assistant for this. again, same thing happens. after a few days, location just stops being reported. With Google maps it was instant.

Many people have recommended Life360 app, but I thought I'd check for a good self-hosted one before trying yet another proprietary third party app.

I'm building a video server and keep hitting a hard freeze I can't pin down. I'd appreciate ideas from anyone who knows Arc (DG2) on Linux, VT-d, or older Xeon platforms.

Hardware:

ASUS Z10PE-D8 WS, BIOS 4301 (latest)

2× Xeon E5-2696 v4 (Broadwell-EP, 22 cores each), microcode 0xb000040

2× Sparkle Arc A310 4 GB (4×HDMI "OMNI View", 8086:56a6 rev 05) - both on CPU 2's slots, PCIe Gen3 x8 and each card has its own PCIe switch in front of the GPU (8086:4fa1), No Resizable BAR on this board (256 MiB BAR)

6 × Kingston valueram 16GB ddr4 2Rx4 PC4-2133P-RA0-11 ECC RDIMM; memtest clean, EDAC 0

Other PCIE: Intel E810 NIC and a AMD FPGA card

Cooler master 1000w Elite Gold (BRAND NEW)

Software

Ubuntu 24.04, HWE kernel 7.0 (7.0.0-34 and 7.0.0-42 both tested)

i915 driver (xe is loaded but not bound)

Firmware: GuC 70.53.0, HuC 7.10.16, DMC 2.08, GSC DG02_2.2357

intel-media-driver (iHD) 24.1.0, later 25.2.4; FFmpeg with VA-API

VT-d on (translated default domain, lazy flush-queue invalidation), interrupt remapping on (x2APIC)

Workload/operation

6 live 1080p50 encodes split across both cards: H.264, HEVC 8/10-bit, AV1 and VP9. Also tested with 8 (4 per card). Each card draws only 14–18 W, at 56–75 °C.

Issue:

The whole server basically stops at once; every network interface goes silent within 0.2 s.

No panic and no reboot, even with kernel.panic=30 and hardlockup_panic=1. My NMI status never reports.

netconsole receives nothing.

After a reset there are no MCE, BERT, EDAC or AER records, with WHEA on or off.

Its not like it turns off, the server still stays powered, fans running; the Q-Code display shows the same AA POST code so nothings different from normal operation.

The board's CATT (CATERR) LED is on and ERR_CPU is off in every freeze where we looked (10 of them). I'm thinking its something to do with the CPU IERR, which explain why the OS logs nothing? I'm very confused.

Tests:

22 freezes in about total of 113 hours of GPU-encode load. The mean is about 5 h; the range is 24 seconds to 14 hours.

I've done the same load with the encodes moved to the CPU and both GPUs idle: ive done a run for 8 hours and one 24 h run and no freeze issues with them...

The GPU encodes alone (no capture cards, no DPDK) still freeze, so im pretty sure its the A310 encodes

Tried, and it still froze:

Kernel 7.0.0-34 → 7.0.0-42, iHD 24 → 25, newer HuC

i915.enable_dc=0

Every OS power saving off: ASPM, runtime PM, D3cold, C-states deeper than C1, SATA LPM, EEE, EPB

Core C6 off, uncore clock capped, turbo off.

Assumptions went to temperature, cooled my room as much as I could the GPU was at 56 °C and still froze.

A310 links forced to Gen2: 8 h clean, then froze in the repeat

intremap=off (interrupt remapping only)

WHEA off

Encode threads and buffers moved to the GPUs' NUMA node

Could test, but haven't (I've done so much testing so I've just tried to post here if anyone has had similar issues and what was the fix):

iommu=pt (GPUs untranslated), or iommu.strict=1 (no lazy flush queue)

The xe driver instead of i915

GSC firmware 2359 (how do you update it on Linux?)

A power limit on the cards through i915's hwmon

A different PSU, or one GPU per CPU socket

My guess:

Both GPUs sit behind the same VT-d unit: the one for CPU 2's root ports, which also covers the E810. My top theory is that two GPUs' constant map/unmap traffic floods that unit with page lookups and invalidations and wedges CPU 2's I/O hub. That would give an IERR with no PCIe error logged.

Or maybe:

cross-socket posted-write credit loss (Broadwell-EP erratum BDF83);

GPU power-state transitions on a lightly loaded card while DMA is in flight.

I've had endless hours and considering just returning the csrds, not sure if return is possible with Overclockers UK ... Alternatively I'll have to have a chat with the manufacturers to get some type of support/repair

Just wondering if anyone else had these issues?

CATERR/IERR hard locks with Arc (DG2) on Broadwell-EP or other older Xeons?

Are there known issues with DG2 under VT-d translation, or with two DG2 cards behind one IOMMU unit?

Is xe on DG2 mature enough for VA-API encodes (HuC/BRC) to be worth testing?

No BMC is fitted (currently shipping from china, but not exactly sure it will be able to identify my exact issue).

Is there any way to capture CPU state at an IERR, such as reading the MCA banks after a warm reset before the BIOS clears them?

What would you try first?

I can post lspci -vvv, dmesg or the kernel config if that helps.

Tldr: I'm stuck, just like my server after any GPU encodes.

I replaced almost all services with selfhosted solutions, but I really struggle to find gmail replacements.

Recently I saw some interesting projects like plmail or mailflow. But I'm kind of reluctant due to the vibe coded nature of such projects.

Some features I'd like to have: webmail and Android app (pwa or native), usable search, snooze. Easy in and export... I'm not switching to proton to be locked in the next golden cage.

Any opinions on mail?

edit: great confusion. I'm not looking for a Mail server, just a client

Anybody have any ideas for replacing the loss of Reddit's RSS feeds? I've probably been using the front page RSS feed for over a decade to get reddit posts. On my phone I like to check specific subreddits, but on my desktop I like to just have an unfiltered firehose of everything hitting /r/all. (http://www.reddit.com/.rss)

Is it possible to replicate that? I briefly did some looking. There's an open-source project called redlib - the main project seems nearly dead and while there are forks, nothing seems to be consistently functional these days.

I don’t care about discord servers. This is not for the public. I just want to use something that has good screen sharing (like discord’s), call support, and a chat.

Calling and chat options are the easy ones to find. I’ve been having trouble finding something for responsive screen sharing. Especially ones that works for group calls.

I will only be using this with people I know. As much as I hate discord I haven’t found a suitable replacement yet.

In summary, key criteria I want are:
- calling
- text messaging
- responsive screen sharing
- group calls

What have people been using for this?

If you have anything with SSH exposed that comes from Docker, alpine, or even generally you should double check you aren't vulnerable:

https://sshhole.com/

Disclosure first: I'm the founder of a technical due diligence tool that reads git histories, which is how this list came about. You don't need it for anything here, and there's no link to it.

Across the 1,577 public projects it has read, it flagged apps that report home by default. These 14 were each checked against the project's own code, and all were still on by default in the latest code on 5 November.

App	Reports to	Turn it off
Grafana	stats.grafana.org	reporting_enabled = false in [analytics]
Jenkins	usage.jenkins.io, from the admin's browser	-Dhudson.model.UsageStatistics.disabled=true
NocoDB	telemetry.nocodb.com	NC_DISABLE_TELE=true
Plane	telemetry.plane.so	the telemetry box at instance setup (is_telemetry_enabled)
Directus	telemetry.directus.io	TELEMETRY=false
SeaweedFS	telemetry.seaweedfs.com	-telemetry=false on the master
Leantime	telemetry.leantime.io	LEAN_ALLOW_TELEMETRY=false
OpnForm	telemetry.opnform.com	OPNFORM_ANONYMOUS_TELEMETRY_DISABLED=true
Checkmate	checkmate-telemetry.bluewave-labs.workers.dev	TELEMETRY_ENABLED=false
Onyx	telemetry.onyx.app	DISABLE_TELEMETRY=true
Coder	telemetry.coder.com	CODER_TELEMETRY_ENABLE=false
Qdrant	telemetry.qdrant.io	telemetry_disabled: true
Saleor	usage-telemetry.saleor.io	SEND_USAGE_TELEMETRY=False
Payload	telemetry.payloadcms.com	telemetry: false in the config

Three that look as if they belong here and don't:

• Syncthing asks before it sends usage reports. Crash reports are on by default (crashReportingEnabled).
• Notifuse's setup wizard shows the telemetry box, ticked, so you see the choice.
• ownCloud's sample config has telemetry.enabled true, but it says community servers without a licence key never send.

This isn't a complaint. Several describe what they send as anonymous, and I haven't judged what each one sends, only that it's on until you say otherwise. The list is what's in my library, not every app. If I've got one wrong, tell me and I'll fix it.

If you'd like analysis and might want to use my tool, DM me.

Heads up if you run Ubuntu or Debian servers. A new APT flaw was published on Nov 4, and how you install the fix matters.

CVE-2026-109518 is in APT's HTTP backend, so it affects apt, apt-get and aptitude. Ubuntu rates it High (CVSS 4.0 score 7.7). An attacker who can tamper with your HTTP traffic, or a malicious mirror, can get APT to install a package that failed signature verification. Package scripts run as root, so that means root on the box. It needs a retried download to work. Neither Ubuntu nor Debian says it is being exploited.

Ubuntu's recommended update turns retries off while you patch:

sudo apt update && sudo apt upgrade -o Acquire::Retries=0

If a full upgrade isn't possible right now:

sudo apt update && sudo apt install -o Acquire::Retries=0 --only-upgrade '?source-package(^apt$)'

To check your version:

apt list --installed '?source-package(^apt$)'

Fixed versions:

- Ubuntu 22.04: 2.4.15ubuntu0.1

- Ubuntu 24.04: 2.8.4ubuntu0.1

- Ubuntu 26.04: 3.2.1ubuntu0.1

- Debian trixie: 3.0.4 (the advisory only names trixie, so check the Debian tracker for other releases)

Other things Ubuntu notes: HTTPS protects against the on-path case, but some community and country mirrors don't support it. Unattended-upgrades applies the patch within 24 hours if you have it enabled.

Question for people running apt-cacher-ng or a local mirror: is the traffic between your clients and the cache plain HTTP, and are you changing anything because of this? Neither advisory covers that setup.

Sources:

- Ubuntu: https://ubuntu.com/blog/apt-vulnerability-fixes-available

- Debian DSA 6607-1: https://lists.debian.org/debian-security-announce/2026/msg00526.html

- Debian tracker: https://security-tracker.debian.org/tracker/apt

(Drafted with AI help. The facts are from the pages linked above.)

Immich Machine Learning (ghcr.io/immich-app/immich-machine-learning:v3) uses a library called ONNX Runtime, from Microsoft.

This library has telemetry enabled by default, and the Docker build of Immich Machine Learning does not disable it.

I found out about this because my DNS logs were showing thousands and thousands of blocked queries to resolve mobile.events.data.microsoft.com.

I don't think this is acceptable for something that is going to process all my photos, especially since it goes to Microsoft, even if it's anonymised or whatever. The whole point of self hosting is avoiding shit like this.

Mitigations:

Add the environment variable ORT_DISABLE_TELEMETRY=1 to the immich-machine-learning container.

Aditionally, it is good practice to isolate all containers that do not need internet access from the internet. For example, with Docker compose, you can create an isolated network for Immich:

networks: immich-iso: internal: true driver: bridge 

Then, on all immich containers:

networks: - immich-iso 

Be sure to add your reverse proxy to this network so that it can access the Immich server.

This was reported to the Immich issue tracker in August and they have not addressed it yet.

Found an old Asus prebuilt in a closet and wondering if it makes sense to use as a starter box or not.

Specs:

• CPU: Intel i5-3340S (4c/4t, 3rd gen, 65W TDP)
• RAM: 4GB DDR3
• GPU: GT 620 2GB
• Storage: 500GB HDD
• PSU: 300W OEM

Is this worth tinkering with, or is 3rd gen Intel idle power so bad that I’d spend more on electricity than just buying a cheap N100 mini PC? Scrap it or keep it?

Hey everyone,

I've been doing some research as to what it takes to move off iCloud as I want to gain some autonomy moving between different desktop and mobile devices. Consequently, this means I'll have to take my family with me for most things.

Some of the candidates I've identified:

1. Passwords -> Vaultwarden
   
2. Photos -> Immich
   
3. Drive -> Seafile
   
4. Calendar -> Davis
   
5. Notes -> Memos
   
6. Notes/groceries -> KitchenOwl
   
7. Hide My Email -> Addy/Proton

I really wanted Immich to be the first as it would cut my iCloud storage bill, but we rely heavily on some of the shared libraries features. I looked into the Noodle Gallery fork, but I'm not quite ready to take that risk yet. So I started with Vaultwarden instead. I documented it in a YouTube video and a written guide with all the commands.

Some of my learnings:

- Switching was really easy: On the iPhone, "Export Data to Another App" moved passwords, passkeys and 2FA codes into the Bitwarden app effortlessly. The awesome CXP protocol makes this all possible.

- Family support isn't that hard either: I turned off sign-ups, created a Family organization, and moved shared logins into the shared collection. An email relay/server is not necessary as they can register with the invited address directly.

- Backups: I made sure I had a proper backup system in place first. I use Proxmox Backup Server.

- Updates: I used the Alpine version of the PVE helper script with automatic updates that run weekly.

- Remote access: By far the most intensive thing to set up. Technically not required, but this allows updates to your vault outside of the house. I initially set it up to use Tailscale with split-horizon DNS. The trade-off is that you need to have Tailscale active on your phone. There are tons of alternatives for this and I'll be tweaking this going forward. Maybe a move to Netbird or open access with better firewall rules.

If you're in the process of moving off iCloud or Google: I'd love to hear about your experiences and which services were the hardest to get the family on board with.