[dev.icinga.com #11072] LWRP icinga2_service is very slow

[icinga-migration]

trafficstars

This issue has been migrated from Redmine: https://dev.icinga.com/issues/11072

Created by ascopenco on 2016-02-02 20:27:38 +00:00

Assignee: vkhatri Status: Assigned Target Version: (none) Last Update: 2016-02-16 09:15:18 +00:00 (in Redmine)

---

Hello,

I have issue with generation icinga2 configs with cookbook LWRPs. It take huge time. For example generation of ~2k services take more then 30 mins on VM with 4 cpu core and 16Gb memory. Tests show me the main problems is in icinga2_service. Its the slowest LWRP. 1-3 secs for one execution.

[icinga-migration]

Updated by ascopenco on 2016-02-02 20:50:34 +00:00

Let me explain how time of execution is growing. At the beginning icinga2_service LWRP executes fast, execution time grow up to 10-15 sec in the end.

[icinga-migration]

Updated by vkhatri on 2016-02-13 07:19:44 +00:00

• Status changed from New to Assigned
• Assigned to set to vkhatri

@ascopenco Sorry for the delay, i will take a look at it. 4cores machine indeed should not take 30mins for ~2k services.

Could you please share the Chef Server and Client details? Would be great to have a LWRP resource sample.

[icinga-migration]

Updated by ascopenco on 2016-02-15 08:13:44 +00:00

Chef Server: 12.3.1 Chef Client: 12.6.0

example of LWRP:

icinga2_service "#{server['fqdn']}_ssh" do import 'check-service-tmpl-30s' display_name 'ssh' host_name server['fqdn'] max_check_attempts 4 check_command 'ssh' custom_vars :notification => notify_list end

[icinga-migration]

Updated by vkhatri on 2016-02-15 13:14:06 +00:00

@ascopenco thanks!

[icinga-migration]

Updated by ascopenco on 2016-02-16 09:15:18 +00:00

in addition:

VM is on CentOS release 6.7 (Final)

strace show lags on brk syscall: [pid 9791] 1.919851 brk(0x15000e000) = 0x15000e000 [pid 9791] 0.001501 brk(0x150032000) = 0x150032000 [pid 9791] 0.001288 brk(0x150056000) = 0x150056000 [pid 9791] 0.001319 brk(0x15024a000) = 0x15024a000 [pid 9791] 0.001286 brk(0x15026e000) = 0x15026e000 ... [pid 9791] 8.383971 brk(0x150421000) = 0x150421000

[vkhatri]

@scopenco Are you still facing the slow start up issue? Never got a chance to get back to you.

[scopenco]

Yes, and I know the reasons. It's O(n*n) algorithm complexity. Each declaration of icinga2_service cause chef-client to parse service.conf, construct hash to services and search declared service in it. So when you have 10 services, you have 10*10 time parser service.conf, then you have 3-5k service, it became 4-6 hours calculations.

So How did I resolved this problem for >3k service installations? I've created a separate resource for icinga2 services declaration. It creates one file per host services so my custom icinga2_service each time read only file with one host with 5-7 services. It works much faster of course.

[olebel]

got same problem in configuration with 2k hosts and 20k services. It just not working because of described reason.

After some experiments - ended up with following approach:

• use cookbook only for Icinga2 packages/main config/endpoints/apilisteners.
• monkey-patched zone resource to get it work and use zone resource from cookbook.
• keep all Icinga configuration items in cookbook files in native Icinga2 syntax.

Reason: Icinga2 configuration has it's own syntax, that's already "monitoring as code". No reason to introduce one more translation layer ( Chef Resources > Icinga2 objects ).

[scopenco]

I've created a new cookbook that creates hosts and services using API https://supermarket.chef.io/cookbooks/icinga2_api so now using this cookbook a get in 10 times better performance.