4 ms·
This is really good - +1 for the shell based installer and time sorting, and a level of polish that infrastructure tools often lack. I'd add (in addition to wh
by nailer 13y ago
This is really good - +1 for the shell based installer and time sorting, and a level of polish that infrastructure tools often lack.
I'd add (in addition to what you have now) your own procfs tools to extend some of the knowledge the older command line tools.
- For a C app you care about argv[0] that top batch mode displays. But for some shell, python or perl or ruby you care about the actual script name. And for Java you care about the .jar (especially in enterprise envs). Even in your example, it's just some anonymous java process. There could be 12 'java's - I want to know whether it's our app or some Enterprise Crapware agent.
- Likewise load average - for a multicore system a load average of 10 could be quite underutilized - if I remember correctly, the run queue is actually (number of logical cores) long.
There's also some things you audience might not be familiar with, but would find valuable - tools like this can be a good way to show off some of this stuff.
- Do you have ethtool in there? Particularly patch disconnect events.
- You can also ask the switch what ports the host is connected to with a two liner LLDP/CDP thing.
These kinds of things are GOLD to people that know they exist, but as you know, Unix tools aren't very discoverable - your app could help with that.
- Not sure if this is relevant, but if you do monitor network stuff, try and steer users away from ifconfig, it's not maintained and will simply skip certain interfaces (eg, trading arbitrage desks used vlans on top of virtual IPs on top of bonds, these are completely invisible to ifconfig. There are problably other cases I haven't run into yet).
Are you guys on Twitter? I'm also working in similar space (used to work at Red Hat and IBM as a Linux specialist, now I write node apps all day). Not competing, but doing similar stuff.
- X4 13y agoGood input, yeah after seeing the demo I've also noticed that the anomaly detection monitoring cannot be solved by simply pointing to a process. I'd suggest aggregating the data of vmstat,ps,lsof,nc etc. into a simplified and abstract summary by incorparating a structure based AI model, like HTM. You could build rules and action lists and anomaly type lists, then let the AI categorize anomalies by pattern and structure detection. Heck you could go much further, ie. building an NLP query interface.. Don't get me wrong, I don't want to smoke your brain, it's just what I would do. Keep working on it!
- nailer 13y ago+1 there's a lot of stuff you can find by correlating between the tools. Since many of these tools are just /procfs readers and you can make something tighter by hitting their data sources directly, rather than scraping existing command line apps (and the source for these apps is generally quite readable).
- treskot 13y agoThanks for your great feedback. We wanted to build a framework which can be easily extendable by adding custom commands. This is just the beginning, we will soon roll out more tools which will help integrate tightly with your server and applications. We plan to integrate monitoring and many other related features after analyzing how the users are using our system. Regarding monitoring network layer, you can always write your own bash which records network activity. Interesting thoughts, we have added them in our pipeline. We are @SeaLionTeam on twitter, we would love further inputs from you
- nailer 13y agoI get what you mean re: letting users add their own tools, but I think there's some utility from providing/suggesting things people don't already know. Eg: a cable gets pulled: - syslog, which is well known, might show a NIC event. - ethtool and openlldp will tell you your were connected to row 4 port 26 of MySwitch and now you're not. Which is super useful when you need a NOC person to go investigate.