4 ms·

Lets make a program that can print strings and do log rotation. Then overly complicate it, put everything in it. wcgw?

classified 5y ago
But I want my logger to brew me a coffee via the network card in my coffee machine. A logger that's not Turing-complete is definitely under-featured.
- nottorp 5y ago
  That's what we're missing. Anyone knows log4j enough to write a proof that it's Turing complete? How about it passing the Turing test?
- stjohnswarts 5y ago
  I want my logger accessible/controlled from my nintendo switch so I can check logs while I game.
christophilus 5y ago
You just need fast print strings. Leave log rotation up to the OS and it’s tools.
- vbezhenar 5y ago
  How does it work? Do I need to close/reopen file after every log message?
  - darthShadow 5y ago
    No, just close and re-open the file when you receive a specific signal, generally SIGHUP, from logrotate or any equivalent tool.
    - vbezhenar 5y ago
      I don't think that Java can handle UNIX signals, at least without some internal API.
- throwaway984393 5y ago
  I wish I could tell you how many times I've seen a machine run out of disk space because logrotate had been failing since the last time someone pushed out a config, or a new log file appeared with a software upgrade. It's gotten to the point where I give /var/log its own disk partition now, and hunt down any logs in /opt and bind mount them into /var/log. Screw the logs, I want my system to keep running. I am 1000% in favor of apps managing their own logs. Generic system agents can collect and ship arbitrary logs remotely for storage/processing.
  - marcosdumay 5y ago
    That's why you set your OS to rotate everything inside /var/log. And yes, chase down the other logs and place them there.
  - quesera 5y ago
    > It's gotten to the point where I give /var/log its own disk partition now, Putting /var on its own partition has always been recommended. When default installer/partitioners fail to do so, I assume it's a habit picked up from desktop-oriented Linux distros. Logging to /opt would clearly be a bug of the brown M&M variety -- it would make me question other decisions made by the developers.
    - lanstin 5y ago
      If you have a standard cloud image and want it to use all the disk given to it, one big xfs partition is easier than a bunch of various little partitions for the sake of not filling the root partition. Hopefully everything log rotates well and you have disk full metrics/alarms when they don't. And the architecture should be if the node goes bad for any reason just replace it (with hysteresis so you don't kill all the zookeepers at once when new traffic and bad rotation/rentention fills the disks all at once)
      - quesera 5y ago
        Your "easier" has a lot of complicating qualifications. :) I'd describe single-partition as "slightly simpler, but requiring additional support mitigations, and less effective at the primary job of keeping systems running". I prefer: One partition for / which is required to have an operable system for diagnosis of problems One partition for /var which is expected to grow with limited predictability One partition for the rest, which should be largely static depending on function, but sometimes surprises you regardless. ... With elastic scaling compute nodes and centralized logging, most of these issues are subsumed into the enormous support infrastructure and can be ignored. But some applications don't map well to that environment.
marcosdumay 5y ago
Well, TIL that log4j does log rotation by itself. Really, why the hell does an application need to rotate logs by itself? It can only lead to bad decisions. Now I understand why java servers never seem to have all the information about bugs on their logs. And looking at the documentation, it has a very Java idea of rotation, where it supports every possible use case as a main case, and the examples expect that you will set it with the most insane defaults from day zero.

But I want my logger to brew me a coffee via the network card in my coffee machine. A logger that's not Turing-complete is definitely under-featured.

nottorp 5y ago
That's what we're missing. Anyone knows log4j enough to write a proof that it's Turing complete? How about it passing the Turing test?
stjohnswarts 5y ago
I want my logger accessible/controlled from my nintendo switch so I can check logs while I game.

christophilus 5y ago

You just need fast print strings. Leave log rotation up to the OS and it’s tools.

vbezhenar 5y ago
How does it work? Do I need to close/reopen file after every log message?
- darthShadow 5y ago
  No, just close and re-open the file when you receive a specific signal, generally SIGHUP, from logrotate or any equivalent tool.
  - vbezhenar 5y ago
    I don't think that Java can handle UNIX signals, at least without some internal API.
throwaway984393 5y ago
I wish I could tell you how many times I've seen a machine run out of disk space because logrotate had been failing since the last time someone pushed out a config, or a new log file appeared with a software upgrade. It's gotten to the point where I give /var/log its own disk partition now, and hunt down any logs in /opt and bind mount them into /var/log. Screw the logs, I want my system to keep running. I am 1000% in favor of apps managing their own logs. Generic system agents can collect and ship arbitrary logs remotely for storage/processing.
- marcosdumay 5y ago
  That's why you set your OS to rotate everything inside /var/log. And yes, chase down the other logs and place them there.
- quesera 5y ago
  > It's gotten to the point where I give /var/log its own disk partition now, Putting /var on its own partition has always been recommended. When default installer/partitioners fail to do so, I assume it's a habit picked up from desktop-oriented Linux distros. Logging to /opt would clearly be a bug of the brown M&M variety -- it would make me question other decisions made by the developers.
  - lanstin 5y ago
    If you have a standard cloud image and want it to use all the disk given to it, one big xfs partition is easier than a bunch of various little partitions for the sake of not filling the root partition. Hopefully everything log rotates well and you have disk full metrics/alarms when they don't. And the architecture should be if the node goes bad for any reason just replace it (with hysteresis so you don't kill all the zookeepers at once when new traffic and bad rotation/rentention fills the disks all at once)
    - quesera 5y ago
      Your "easier" has a lot of complicating qualifications. :) I'd describe single-partition as "slightly simpler, but requiring additional support mitigations, and less effective at the primary job of keeping systems running". I prefer: One partition for / which is required to have an operable system for diagnosis of problems One partition for /var which is expected to grow with limited predictability One partition for the rest, which should be largely static depending on function, but sometimes surprises you regardless. ... With elastic scaling compute nodes and centralized logging, most of these issues are subsumed into the enormous support infrastructure and can be ignored. But some applications don't map well to that environment.

marcosdumay 5y ago

Well, TIL that log4j does log rotation by itself. Really, why the hell does an application need to rotate logs by itself? It can only lead to bad decisions. Now I understand why java servers never seem to have all the information about bugs on their logs. And looking at the documentation, it has a very Java idea of rotation, where it supports every possible use case as a main case, and the examples expect that you will set it with the most insane defaults from day zero.