collectd plugin ordering matters. the write_graphite plugin establishes a TCP connection but there is a race between that and the df/du values that get reported. du is especially problematic since we report this only every 12 hours. so, instead we cache the values and report it every 20 seconds. on the carbon side, it will just retain every 12 hours (since that is the whisper retention period). there is also FlushInterval which I am not 100% sure has any effect. by default, the write_graphite plugin waits for 1428 bytes to be accumulated. (https://manpages.debian.org/unstable/collectd-core/collectd.conf.5.en.html) https://github.com/collectd/collectd/issues/2672 https://github.com/collectd/collectd/pull/1044 I found this syntax hidden deep inside https://www.cisco.com/c/en/us/td/docs/net_mgmt/virtual_topology_system/2_6_3/user_guide/Cisco_VTS_2_6_3_User_Guide/Cisco_VTS_2_6_1_User_Guide_chapter_01111.pdf