Hi, No leaks. The VM has 1GB, very light for moderate busy relay. Should be 4GB, perhaps 3DB would do. Settings should include: AvoidDiskWrites 1 DisableAllSwap 1 MaxMemInQueues 1024MB # perhaps 1536MB My relays run about 800MB to 1.2GB. A nice way to view memory usage is with egrep '^Vm' /proc/$(cat /var/run/tor.pid)/status which might show something like VmPeak: 871696 kB VmSize: 710152 kB VmLck: 710148 kB VmHWM: 810872 kB VmRSS: 649328 kB VmData: 492052 kB VmStk: 93 kB VmExe: 4704 kB VmLib: 104724 kB VmPTE: 1044 kB Notice that current VmSize usage is less than VmPeak, indicating that memory was release by glibc malloc. Tor consumes substantial skbuf memory in the kernel, which accounts for some of the difference in reported size for VmRSS and VmSize and total memory consumption.
On Mon, 24 Jun 2019 01:49:18 -0400 starlight.2018q2@binnacle.cx wrote:
Tor consumes substantial skbuf memory in the kernel, which accounts for some of the difference in reported size for VmRSS and VmSize and total memory consumption.
This reminded me that I had these tweaks applied (as on all my boxes): net.core.rmem_max=16777216 net.core.wmem_max=16777216 net.ipv4.tcp_rmem=4096 524288 16777216 net.ipv4.tcp_wmem=4096 524288 16777216 Maybe these are too much for 1GB VMs with 2 instances of Tor each, so now switched it all back to defaults (which didn't seem to affect Tor speeds) and it seems better so far. -- With respect, Roman
On 24 Jun (01:49:18), starlight.2018q2@binnacle.cx wrote:
Hi,
No leaks. The VM has 1GB, very light for moderate busy relay.
Should be 4GB, perhaps 3DB would do. Settings should include:
AvoidDiskWrites 1 DisableAllSwap 1 MaxMemInQueues 1024MB # perhaps 1536MB
My relays run about 800MB to 1.2GB. A nice way to view memory usage is with
egrep '^Vm' /proc/$(cat /var/run/tor.pid)/status
which might show something like
VmPeak: 871696 kB VmSize: 710152 kB VmLck: 710148 kB VmHWM: 810872 kB VmRSS: 649328 kB VmData: 492052 kB VmStk: 93 kB VmExe: 4704 kB VmLib: 104724 kB VmPTE: 1044 kB
Notice that current VmSize usage is less than VmPeak, indicating that memory was release by glibc malloc.
Tor consumes substantial skbuf memory in the kernel, which accounts for some of the difference in reported size for VmRSS and VmSize and total memory consumption.
What OS is this running on? Can you possibly send back a SIGUSR1 log? Kernel memory leak means we are buffer bloating it with cells... which we aren't suppose to do these days :S ... Thanks! David -- mtl83aVORGbFMy0uOJnLcNQG64cDwnBUp5hK73imIrw=
participants (3)
-
David Goulet -
Roman Mamedov -
starlight.2018q2＠binnacle.cx