all: reduce MTR report cycle from 10 to 1

We have received a complaint from InnovaHosting about them being hit by
about 150 ICMP `ttl1` packets/s on their routers, causing excess CPU usage.
https://client.innovahosting.net/viewticket.php?tid=532874&c=8gALx9vm

By using `tcpdump` I have identified that `mtr` by default pings the
target 10 times, which means that the default value of `-c`/`--report-cycles`
is 10, although this is not documented in the manual.

We can see this when calling `mtr github.com` and watching with `tcpdump`:
```
 > sudo tcpdump -v -i eno1 icmp and src 185.181.230.78 and dst github.com | grep 'ttl 1,'
tcpdump: listening on eno1, link-type EN10MB (Ethernet), snapshot length 262144 bytes
19:54:53.981243 IP (tos 0x0, ttl 1, id 37119, offset 0, flags [none], proto ICMP (1), length 64)
...(8 packets omitted)...
19:55:03.025460 IP (tos 0x0, ttl 1, id 38226, offset 0, flags [none], proto ICMP (1), length 64)
```
We don't need to run the test 10 times to get a result for our metric.

Related to:
https://github.com/status-im/infra-role-bootstrap-linux/commit/ea22bdfe

Signed-off-by: Jakub Sokołowski <jakub@status.im>

This commit is contained in:

Jakub Sokołowski

2023-11-20 21:04:59 +01:00

parent 6b800a5342

commit 24020d0962

No known key found for this signature in database

GPG Key ID: FE65CD384D5BF7B4

1 changed files with 4 additions and 0 deletions

									
										4

ansible/group_vars/all.yml

										View File
									
				@ -50,6 +50,10 @@ bootstrap_sysctl_config:

				  # Allow calling 'perf' without root'

				  kernel.perf_event_paranoid: 1

				# MTR network latency metrics

				bootstrap__mtr_jobs:

				  - { name: 'github-icmp', addr: 'github.com', flags: '-c1' }

				# Consul Catalog Query URL

				consul_catalog_url: 'http://localhost:8500/v1/catalog'

all: reduce MTR report cycle from 10 to 1

4 ansible/group_vars/all.yml Unescape Escape View File

4

ansible/group_vars/all.yml

View File