bioe007 / synopsys

synopsys - A SYNOPsis of your SYStem

The goal is to create an overview of a system. Instead of running:

• uptime
• dmesg | tail
• free
• vmstat
• mpstat
• iostat
• sar DEV, TCP, ETCP
• top

Show the critical parts from each of those in legible fashion with a single command.

Ideally this would be doable from a single command that doesn't need to be pre-installed so either a script or self-contained binary.

Current state

This isn't really useable, but it does output:

• running / total processes, number of vcores, current cpu frequency
• load average
• cpus - can show top N cpus sorted by user time
• memory - free/total, buff/cache
• disks - total requests, written/read KB
• uptime

there are also some cli options to limit the number of CPU and disks shown, to show disks only, etc..

Notes

This is just thinking out-loud stuff..

Because any scripted language (e.g. python, lua) doing this would require either a ton of boilerplate or a long list of dependencies that might not be available to a sysadmin during a critical outage I'm favoring something that's easy to build/store/distribute as a single binary so even if it's not resident on a struggling system a simple the tool is only a simple scp synopsys away

Critical Parts

These are the parameters I'd like to show

• Uptime: express as hours:min:sec (done)

• load average (done)

  • todo Enable mode where R/num_vc and B/num_disks ?

• CPU: cores, overall % useage, then % sys, usr, guest, ... (done)

  • TODO Have a mode that shows top 'any%' so if/when CPU are above a threhshold in any usage category they're shown. So if a system has some high guest% cores and some high system% cores all can be shown.
  • by default, show the overall cpu utilization (done)
  • wonders any way to make mpstat type of info here?

• Processes:

  • run|able, sleep, unint sleep, zombies
  • TODO

• 'Errors' from dmesg and ~ dmesg | tail (or journalctl -b | tail)

  • todo

• Memory: free/used (proc/meminfo) (done)

• Swap: free/used (proc/meminfo)

  • todo

• Disk activity: rw/wr in MBs and queue size

  • /proc/diskstats - (done'ish)
  • /proc/partitions - (done)

• Network In/Out (per device?) - TODO

  • connections - active, passive, trans/retrans stats
  • top 'few' processes consuming CPU | memory

Display

Initially this will just output some rolling format. will have to think about something like a tui to properly place things for readability though

Random thoughts

Is there a faster way to fetch all this data than reading a text file each time?

lots of repetition with a ThingStats that has 'previous', 'current' and does math on all the fields. I could have something like a generic Delta struct that accepts the same type twice, does the math between all fields etc.. But this would require making everything public in the structs.

References

• /proc
• /proc/loadavg
• filesystems
• /sys/block
  • /proc/diskstats