redundant cloud servers

Attilla de Groot attilla at attilla.nl
Sat Apr 24 13:50:19 PDT 2021


Thank you, it works nicely again.

-- Attilla

On April 24, 2021 22:22:47 Dirk Hohndel <dirk at hohndel.org> wrote:

> Yes, the US server VM ran out of memory and killed a process that didn't 
> have an auto-restart in its systemd service file...
>
> Should be up and running again.
>
> Those who used the latest master SHOULD have been automatically redirected 
> to the European server which continued working :-)
>
> /D
>
>> On Apr 24, 2021, at 11:54 AM, Attilla de Groot <attilla at attilla.nl> wrote:
>>
>> Hi Dirk,
>>
>> Are there any issues with the backend today? I’m not able to sync either 
>> from my desktop or mobile (also over 4G).
>>
>>
>> — Attilla
>>
>>> On 19 Apr 2021, at 22:58, Dirk Hohndel via subsurface 
>>> <subsurface at subsurface-divelog.org> wrote:
>>>
>>> Hi Jason,
>>>
>>> Thanks for the feedback. Yes, the new DNS records now have much longer TTL 
>>> and are therefore more likely to be cached.
>>> I also made some subtle changes to the logic when we first attempt to 
>>> connect and actually try to start writing to the cloud that might make a 
>>> difference there.
>>>
>>> /D
>>>
>>>> On Apr 19, 2021, at 1:50 PM, Jason Bramwell  wrote:
>>>>
>>>> Many thanks as always.
>>>>
>>>> I used to see a ‘feature’ where the first attempt to load Subsurface would 
>>>> fall back to using the local cache and i’d have to close/reopen Subsurface 
>>>> to get it to connect to the cloud storage, Linux diagnosed this as likely 
>>>> slow DNS lookup, very brief testing (I do mean very brief) seems to 
>>>> indicate that this is no longer the save now and it ‘just works’.
>>>>
>>>> Jason
>>>>
>>>> Sent from Mail for Windows 10
>>>>
>>>> From: Dirk Hohndel via subsurface
>>>> Sent: 19 April 2021 21:00
>>>> To: Subsurface Mailing List
>>>> Subject: redundant cloud servers
>>>>
>>>>
>>>> Over the past couple of weeks I quietly moved the cloud server backend and 
>>>> the authentication database out of my home network and back into the public 
>>>> cloud.
>>>> The fact that apparently no one noticed and nothing broke was rather 
>>>> encouraging :-)
>>>>
>>>> I then setup a second cloud server in the public cloud, this one in Europe 
>>>> (actually, Frankfurt, Germany).
>>>>
>>>> A few minutes ago I merged a pull request that adds a couple of interesting 
>>>> new features to both Subsurface and Subsurface-mobile:
>>>>
>>>> - geo based choice of cloud server (those in the Americas should end up 
>>>> using the US server, everyone else should get directed to the European 
>>>> server), which is intended to be completely transparent to the user.
>>>> I.e., unless you explicitly look for it, you shouldn't notice (or care) 
>>>> which server you are using
>>>> - hot failover between the two servers (so if the server that you have used 
>>>> before is unavailable, Subsurface and Subsurface-mobile should try the 
>>>> other one - and the implementation makes it easy to add more servers if 
>>>> that appears useful)
>>>>
>>>> The two servers are automatically kept in sync in the background. I can 
>>>> create one extremely artificial case where there is a fairly short 
>>>> (generally sub one second in my tests) window for a race condition. A user 
>>>> would have to have two different devices that are synced to different 
>>>> servers (for example because they took their phone with them on an overseas 
>>>> dive trip), that have made conflicting changes to the dive log, and the 
>>>> user would then have to cause a save to the cloud on both devices within 
>>>> the roughly one second window. Forcing that race turned out to be fairly 
>>>> hard because the mobile app does a few other things before storing the data 
>>>> to the cloud - so for testing I ended up using two instances of Subsurface 
>>>> and synchronized clicks... really -- not a realistic scenario.
>>>>
>>>> If this were to happen, I will get an automated email alert and fixing the 
>>>> problem isn't really hard - but I wanted to give a full disclosure of the 
>>>> issues that I am aware of.
>>>>
>>>> What would help right now if people tested all this.
>>>> So far I have not had the time to make new beta versions of 
>>>> Subsurface-mobile, but as I said, it's actually easier to test on the desktop.
>>>>
>>>> /D
>>>> _______________________________________________
>>>> subsurface mailing list
>>>> subsurface at subsurface-divelog.org
>>>> http://lists.subsurface-divelog.org/cgi-bin/mailman/listinfo/subsurface
>>>
>>> _______________________________________________
>>> subsurface mailing list
>>> subsurface at subsurface-divelog.org
>>> http://lists.subsurface-divelog.org/cgi-bin/mailman/listinfo/subsurface
>>

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.subsurface-divelog.org/pipermail/subsurface/attachments/20210424/d1805abc/attachment.htm>


More information about the subsurface mailing list