6 months with a .tk domain for production.

GuruDeutschlag · 2026-07-06T13:17:24+00:00

Six Months with a.tk Domain for Production Use Yes, I know what you are thinking [b]No, do not use freetlds in production[/b] yes Six months ago I deployed a Microservice cluster on a 512MB Hetzner VP...

kate3

Member

Joined:: Jul 2024
Posts:: 208
From:: Utrecht, NL

1mo ago · edited

#11

GuruDeutschlag said:

Month four was DNS propagation delays yes up to 3600 seconds

3600 seconds is the TTL ceiling on most .tk zones, not propagation lag. The registry caps it deliberately. Your "delay" was probably resolver cache expiry, not upstream push failure. Check with

dig +trace

next time; if the authoritative returns immediately but Cloudflare/Google cache stale, that's on you for not tuning TTL down before the incident.

Still, 400-second failover is respectable for a free TLD. Most people running these things do not even have a second registrar.

virsh list --all | wc -l: 47

minhmin

Member

Uptime Junkie

Joined:: Sep 2024
Posts:: 41
From:: Ho Chi Minh City, Vietnam

1mo ago · edited

#12

kate3 said:

Still, 400-second failover is respectable for a free TLD.

Respectable is 60 seconds. 400 is coffee break territory. My Contabo monitoring page fires at 90 seconds, script pulls DNS from three sources, updates Cloudflare origin if needed. 7 minutes was worst case, not average. Average is 4.

But yes, free TLD, you get what you pay for. I pay for my backup now. .co cheap enough.

Fritz48

Member

Datenschutz First

Joined:: Jul 2024
Posts:: 178
From:: Berlin, DE

1mo ago · edited

#13

minhmin said:

I pay for my backup now. .co cheap enough.

.co is GoDaddy registry, US CLOUD Act exposure. For EU processing you want .eu or a national ccTLD with local registrar and GDPR-model contract. The 3-10 EUR price difference is negligible against an Article 83 fine.

Also: Contabo is German-domiciled, BSI C5 certified. Their object storage in Nuremberg keeps logs inside territorial scope if you configure it so. Worth checking — https://contabo.com — for your monitoring artifacts.

Neuland. Aber schnell.

Joy

Member

Joined:: Jul 2024
Posts:: 108
From:: Surabaya, Indonesia

1mo ago · edited

#14

minhmin said:

Moved to .ml; same script; 7 minute failover

Same thing happened to me, .ml also free, also no renewal notice that arrived yes I check spam folder now every Monday yes my coffee is still 12000 IDR, VPS is 18000 IDR, domain was 0 IDR, now domain is 120000 IDR per year, still cheaper than coffee per month yes

ben20

Member

Joined:: Nov 2024
Posts:: 27
From:: Dublin, Ireland

1mo ago · edited

2 #15

Fritz48 said:

.co is GoDaddy registry, US CLOUD Act exposure.

GoDaddy sold .co to Neustar in 2020, then Neustar registry assets went to GoDaddy Registry as a separate entity, then that got bought by Ethos Capital in 2023. The CLOUD Act exposure is the same regardless, US-based registry. But let us be precise about who runs what.

I read the .ml terms once. Freenom can suspend for "technical reasons" without notice. No arbitration, no refund, no data portability. The fun was not sarcastic, it was despair.

JO

josh13

Member

Joined:: Aug 2024
Posts:: 8
From:: Vienna, Austria

1mo ago · edited

1 #16

ben20 said:

https://github.com/gurudeutschlag/tk-failover

Looked at the script. You are using

dig @1.1.1.1

as health check, then falling back to

@8.8.8.8

. Both anycast, both can cache the same stale record. Add

@a.dnspod.com

or

@dns.google

with

+norecurse

to hit authoritative directly. Otherwise you are measuring CDN cache, not registry state.

Also Hetzner Falkenstein to Nuremberg is 5ms, your 400s failover is not network latency, it is script sleep interval. Fix the poll frequency.

GuruDeutschlag

Member OP

Joined:: Jul 2024
Posts:: 122
From:: Munich, Germany

1mo ago · edited

4 #17

josh13 said:

Also Hetzner Falkenstein to Nuremberg is 5ms, your 400s failover is not network latency, it is script sleep interval.

Thank you yes implemented yes the 400 seconds includes my conservative sleep and three retry loops yes I was more afraid of false positive failover than slow failover yes

Also the Hetzner VPS is in Falkenstein yes the Contabo monitoring is on their Singapore instance yes 170ms away yes the 7 minutes minhmin reports is probably similar geography yes

JokoNord

Member

Joined:: Jun 2024
Posts:: 190
From:: Jakarta, ID

1mo ago · edited

#18

GuruDeutschlag said:

The 7 minutes minhmin reports is probably similar geography yes

Sir, very good analysis sir. I use RackNerd Los Angeles, very far from my Jakarta customers sir. But price is good sir. I think I will study your script and make for my .id domains sir, .id not free but stable sir, worth it for customer trust sir. Thank you very much for sharing sir.

traffic worse than my packet loss

YuriDavid

Member

Joined:: May 2024
Posts:: 174
From:: Kyiv, Ukraine

1mo ago · edited

6 #19

JokoNord said:

.id not free but stable sir

.id is PANDI, local Indonesian registry. They had a 3-day outage in 2021, DNSSEC key rollover went wrong. Not free, not immune. The only stable domain is the one you have tested failover for.

Also: spaces. Tabs are for people who hate their colleagues.

indentation is not optional

zoekyo

Member

Less Is More

Joined:: May 2024
Posts:: 102
From:: Kyoto, JP

1mo ago · edited

#20

YuriDavid said:

The only stable domain is the one you have tested failover for.

This. I run one small server in Tokyo, one domain, paid, no failover script. If it breaks I fix it. The alert is my users emailing me. Less is more. The overhead of monitoring three free registries across two continents sounds like a second job.

Guru deutschlag, your efficiency is impressive but I wonder at the cost in attention.

rm -rf / --no-preserve-root ☯

Post a reply