OpenSpace session recording "save frames" crash

On my Lenovo ThinkCentre M93 machine with an NVidia GTX1050 graphics card, OpenSpace session recording playback with "save frames" was crashing after a few hundred frames being saved. Each frame was taking around 1000 to 1500 ms for 4096x4096 frames.

According to Claude.ai, the crashes might be due to CEF "being starved" between updates - and I submitted this fix as an issue on the OpenSpace repo.

Claude suggested workaround was in SGCT.

// sgct/src/image.cpp

png_set_compression_level(png, -1);  // changed to:

png_set_compression_level(png, 1);

As the comment just above that line says, -1 sets the compression level to default, which seems to be 6. Setting this to 1 made the frame export only 300 ms per frame (3x faster) and also fixed the crashes - so CEF was not "being starved" of updates.

The increase in file size was not much, 4 MB increased to 5.1 MB for a typical frame.

Another change I made before exporting again, was to close all other windows and apps like the Edge Browser, which I had used earlier for triggering the exports - now OpenSpace has some checkboxes in the export settings which allows us to hide the UI while exporting, so we don't need to trigger the export via a web browser.

Edit: No, the problem is not solved. No crash when saving frames for this recording, ts_003_moon_to_Earth.osrectxt

but crashes after a few hundred frames when saving frames for this recording, ts_002_india_rotate_zoom.osrectxt

So my current workaround was to split the recording into multiple smaller recordings.

Edit - another workaround to get around the crashes, which seem to be caused by CEF, seems to be to disable the module CEFui inside OpenSpace > Settings - issue thread is at https://github.com/OpenSpace/OpenSpace/issues/3954 

uninstalling ollama

https://github.com/ollama/ollama/issues/986

ran all the lines suggested there, got some 15 GB freed.

Copy-pasting from the above,

Here's a general guideline on how to uninstall it:

Delete the Ollama binary:

Use the rm command to remove the Ollama binary. For example:

sudo rm /usr/local/bin/ollama

If the script created a systemd service, disable and remove it:

If the script created a systemd service for Ollama, you should disable and remove it using the following commands:

sudo systemctl stop ollama

sudo systemctl disable ollama

sudo rm /etc/systemd/system/ollama.service

sudo systemctl daemon-reload

Remove any created user and group (if applicable):

The script might have created a user and group named "ollama." You can remove them using the following commands:

sudo userdel ollama

sudo groupdel ollama

-----------------------

More lines from other people, esp if location of ollama and downloaded models are different - 

sudo rm -v /usr/local/bin/ollama

sudo rm -rv /usr/local/lib/ollama

sudo systemctl stop ollama

sudo systemctl disable ollama

sudo rm -v /etc/systemd/system/ollama.service

sudo systemctl daemon-reload

sudo userdel ollama

sudo groupdel ollama

sudo rm -rv /usr/share/ollama

# for downloaded models

rm -rv ~/.ollama

sudo rm -rv /usr/lib/ollama/

sudo rm -rv /usr/local/lib/ollama

domain transfers - godaddy to nettigritty and one account to another

We had a few domains which needed to be transferred. A couple of domains needed to be transferred from one account to another on nettigritty, and a domain needed to be transferred from godaddy to nettigritty.

downloading from a server via http (no SSL) with auth

Downloading some planetarium files from a server in Sweden, which had password protected directories but no SSL (no https) - 

1. Directly downloading via the browser - slow, prompts us to "Keep" or delete the file every time due to being downloaded unsecurely
   
   
2. Using aria2 - 
   
    aria2c -x16 -s16 --http-user=USERNAME --http-passwd=PASSWORD url
   
   Need ProtonVPN, since without it, http downloads are severely throttled to less than 100 kbps.
   
   
3. Better option, running 
   
   aria2c -x16 -s16 --http-user=USERNAME --http-passwd=PASSWORD -i urls.txt
   
   on one of our AWS servers - all the URLs get downloaded in a jiffy, then we can use Filezilla and download to our local machine over sftp without throttling issues.

change of email id in CMS

One of the users in our CMS (google apps script with php/pgsql authentication) changed their email id in the profile, but was then unable to find any of the assets they had been working on.

Now changed the 'username' field in the postgresql database for this user from

old@gmail to new@new.email

(the email id field had already been changed by the user.)

That seems to have solved the issue. This is a duplicate of an earlier post, https://hnsws.blogspot.com/2024/06/unable-to-see-old-assets-after-email.html

OpenSpace - hover over a planet and allow it to rotate under you - OpenSpace seminar

As noted in an earlier post (which I had forgotten), the relevant option is in Settings > Navigation Handler > "Follow Anchor Node Rotation" 

When I asked Gemini via aistudio.google.com, it found a transcript from an OpenSpace seminar, which is a useful link - https://www.ppadomes.org/events/online-seminars/ppsarchive/2025-archive/2025-01-31-chat

github 14 day traffic stats

Since github has a page which shows 14-day stats, like https://github.com/hn-88/OpenSpace-AppImage/graphs/traffic but which is behind the login of the user, I asked Claude to create a stats page for the traffic to all my public repos. The prompt is here, 

https://github.com/hn-88/hn-88.github.io/blob/master/stats/prompt.md

and the github traffic stats page is here

https://hn-88.github.io/stats/

The "clones" number is so high for the https://github.com/hn-88/OpenSpace-AppImage/ repo because of the multiple daily build workflows.

Enable MFA for Amazon AWS console

PB sent an email asking us to set up MFA for the AWS console logins. I replied that those users don't seem to have the permissions to set up MFA. So, probably set it up using his master login, in IAM, for those users - Google Authenticator. Unfortunately, the name in Google Authenticator gets truncated in portrait mode, so Amazon Web Services is seen, but the account identifying part which comes after is visible only in landscape mode.

adjusting sound volume on Mac

Since I don't have an Apple keyboard, adjusting the volume on the Mac mini was not very straightforward. An LLM's advice seemed to be a bit sarcastic when taken in the context of how easy it should be versus the method suggested - 

Adjusting the volume using the menu bar is a simple and straightforward method. Here are the steps to follow:

1. Click on the Apple logo in the top-left corner of your screen.
2. Select "System Preferences" from the drop-down menu.
3. In the System Preferences window, click on "Sound."
4. Go to the "Output" tab.
5. Drag the volume slider to adjust the volume level.

On my machine, it was "System Settings" and not Preferences. And, there seemed to be a slightly faster way, which is clicking on the "Control Centre" in the menu bar - screenshot below - the "edit controls" button allows us to drag the icon into the menu bar, too.

SSL certificates for an internal web server as backend for app

There was a query about installing SSL certificates for an internal web server, which was to be used as backend for app. Copy-pasting my reply.

Since the app needed trusted certificates, just using a self-signed certificate would be unwieldy, probably needing the installation of our own cert chain on each device or something like that. 

Also, since free SSL certificates from https://letsencrypt.org/ need updating every 60/90 days, buying a certificate may be an option with fewer hassles for this internal server with no external ip address.

https://www.nettigritty.com/sslcertificate.php

has low-cost options. Most probably the lowest cost one will work, can ask nettigritty support if you have doubts.

There are lots of step-by-step tutorials on how to install, like

https://www.geeksforgeeks.org/installation-guide/how-to-install-an-ssl-certificate-on-apache/

(Can find more with https://duckduckgo.com/?q=how+to+installed+a+purchased+ssl+certificate+on+my+apache+server+on+centos+stream&ia=web )

A week later, the reply was,

it is just a java application running with java.

NO Apache. nginix is found. so tried installing with let's encrypt website using ngnix

getting thie error... 

 Detail: DNS problem: NXDOMAIN looking up A for theinternaldomain.net - check that a DNS record exists for this domain; DNS problem: SERVFAIL looking up AAAA for theinternaldomain.net - the domain's nameservers may be malfunctioning

Hint: The Certificate Authority failed to verify the temporary nginx configuration changes made by Certbot. Ensure the listed domains point to this nginx server and that it is accessible from the internet.

My reply was,

You will probably need to buy a certificate and install it, as mentioned in the first email in this thread.

Since theinternaldomain.net is not a publicly available DNS record, certbot cannot resolve it.

If the server is a java application running on some port, say 5000, you will need to run ngnix or apache as a reverse proxy, and install the certificate for apache/ngnix.

The reverse-proxy config for apache is mentioned here, 

https://hnsws.blogspot.com/2021/01/adding-ssl-https-to-http-mp3-streams.html

(but of course, don't use certbot, purchase and install a cert instead as mentioned in the first email in this thread.)

interesting - No-AI chatbot

Interesting to see how chatbots can be programmed manually, without any AI - https://p5js.org/tutorials/criticalai4-no-ai-chatbot/

Via https://p5js.org/

Via https://github.com/hedbergj/sims-dev/blob/gh-pages/expanding-universe-2/index.html

using Blender to mask bright scenes

Using the same technique as in this previous post, but running on Blender 5 on Mac Mini, 3840x2160 video for Unseen Earth was being rendered at around 15 fps using the ProRES LT codec. Apparently Blender still doesn't have GPU accelerated output codec support, and this is one of the fastest codecs as suggested by ChatGPT. The ProRES LT can then be re-encoded as HEVC using the py-ffmpeg-warp presets. 

Some more tweaks used were:

• Multiple masks one atop another to darken extremely bright areas,
• Using Shift-D to duplicate and then K to cut the mask
• Using an adjustment layer to increase contrast for the base layer. 

Edit: April 2026 - I now see that Blender 5 has "Curves" as one of the Adjustment layer modifiers - so, we can easily clamp bright regions using a "Curves" filter. And it looks like this has been there in older Blender versions as well.

close all Finder windows on Mac

It looks like Option+Cmd+w is the keyboard shortcut to close all windows of an app. Also, Finder has this method of Option+click on the File menu, which then shows a "Close all" menu item. 

A bit clunky in my opinion. And apparently many people moving from Linux/Windows to Mac have this problem, due to selecting a large number of files and hitting "Get Info..." like I did - MacOS then proceeds to open that many info boxes instead of showing a single window with cumulative info as on Linux/Windows.

tutorials on claude code - ollama - on Google Colab

 https://github.com/musistudio/claude-code-router - "...route Claude Code requests to different models and customize any request."

running locally - https://www.youtube.com/watch?v=nYDUdnMVDdU

claude code + ollama, another tutorial - https://www.youtube.com/watch?v=3x2q6-5XbQ8

ollama on Google Colab - https://www.youtube.com/watch?v=Y4R-DyvAux8

Moodle custom reports not being emailed

One of our Moodle instances had this issue being reported, that some newly created custom reports were not being emailed as scheduled. 

The issue was that notifications - including email notifications - were disabled for "Custom report builder schedules" at https://ourserver.org/admin/message.php

After enabling email notifications now for this, "Custom report builder schedules" - the issue was reportedly resolved.

DNS resolving glitch with lovable

One of our colleagues wanted a domain verified by "lovable the AI agent" - https://lovable.dev/

After setting up the verification txt record and the provided A record, he still got an error 1000 when trying to browse the website on the newly added subdomain - DNS points to prohibited IP.

Lovable suggested removing the proxy toggle on cloudflare (but that was already set to DNS-only).

So, I suggested that he maybe needs to click on some "verify domain ownership" button somewhere in "lovable" control panel if any ... 

He replied that he had initially used bigrock setup, now he "removed the domain in project settings of loveable and added again, it is now working."

Google Group Admin Settings

 PB noted that he had changed one of our public google group's settings - 

Have made the following changes in the Google Group Admin Settings

Admin Console → Apps → Google Workspace → Groups for Business → Sharing settings

Changed the admin settings from "Private" to "Public on the internet"

This allows to anyone on the web to join the group if the owner of the group has allowed 

Earlier with sharing options was "Private" only the Organisation Members could see the group like:

With this change it is up to the owner of the group to choose one of the following:

This change now allows the gmail users to join the group.

Later, he confirmed that the permission to post is still restricted, only the owner can post:

 

google workspace storage limits

At some point of time, Google Workspace has added a feature to admin.google.com where we can set individual users' storage limits on or off, or change the value on an organizational unit basis. Our current Google workspace now has this feature. One possible catch is that the on/off setting is inherited from the base organizational unit. And it also seems to allow certain users to use 146% of their quota!

pdfs - make smaller - split or join

There was a requirement to 
1. Make some pdfs smaller than 200 kB to enable uploading on a govt. website
2. Split some pages of the pdf into separate documents.

I used pdfsam - https://pdfsam.org/download-pdfsam-basic/ - to do the separation and merging, 

and the online tool

https://www.adobe.com/acrobat/online/compress-pdf.html

to compress the pdf size.

troubleshooting dot net server crash on AWS

First, checked if the AWS dashboard has any troubleshooting info for me. But "This user does not have permissions to view AWS Health events" - so I wouldn't get much help from it. Used my earlier post on investigating another Linux server crash as well as ChatGPT for going step by step.

ChatGPT suggested these,

dmesg -T | grep -i oom

journalctl -k | grep -i oom

which had no results. No out-of-memory events seen with those commands.

du -sh /var/log showed 4 GB of journal

journalctl --disk-usage

also showed nearly 4 GB of current and archived logs.

But the system had enough hard disk space left, 

df -h

showed around 30% remaining for the OS disk. 

Checking authentication log with

tail /var/log/auth.log -n4000 | more

found an entry

2026-01-25T14:38:58.186744+05:30 ip-10-0-0-73 systemd-logind[600]: New seat seat0.

But ChatGPT said that's normal, this is apparently done for each login.

Then, copy-pasting the lines just before the crash from syslog, using T13 since the crash was around 13:30

sudo cat  syslog | grep T13 > /home/theadminaccount/syslogsnippet.txt

ChatGPT immediately found the issue. 

ourtestapi.service: Scheduled restart job, restart counter is at 101203

and the reason the service was failing was

The application '/home/thepath/www/ourapi/ourtest.dll' does not exist.

Stopped the service, disabled it, and deleted it from /etc/systemd/system

sudo systemctl stop ourtestapi

sudo systemctl disable ourtestapi

sudo rm /etc/systemd/system/ourtestapi.service

Additionally, as ChatGPT suggested, set up some limits for journalctl's logging by editing

/etc/systemd/journald.conf

and uncommenting these lines and adding values for them - under

[Journal]

SystemMaxUse=900M

SystemKeepFree=1G

RuntimeMaxUse=100M

MaxRetentionSec=30day

and restarted,

sudo systemctl restart systemd-journald

Also vacuumed (removed archived logs) with

sudo journalctl --vacuum-size=3000M

(0 bytes removed - archives were smaller than that)

sudo journalctl --vacuum-size=300M

(500M of archives were removed.)