ICMP echo requests from another EC2 instance in same availability zone would show packet loss.
We could see it also from multiple monitoring nodes with Pingdom services. The haproxy server itself would not report much errors except problems talking to backends but nothing in dmesg or ethtool. 2013/4/10 Will Glass-Husain <[email protected]> > That seems a pretty good theory. Specifically, how were you testing for > packet loss? Were you using ethtool? > > WILL > > > On Wed, Apr 10, 2013 at 8:20 AM, Maxime Ducharme > <[email protected]>wrote: > >> Check if you have packet loss when this happens, we have seen similar >> issues which were explained by AWS tech support that the IRQ of network >> card of the physical machine was saturated by another customer of AWS. This >> was resosurce issue. >> >> The bigger instance you take, the more resources you will get but it is >> never guaranteed according to AWS guys. >> >> We ended up having ELB in front of multiple big instances (cc types) and >> problem is now resolved. >> >> >> 2013/4/10 Will Glass-Husain <[email protected]> >> >>> That's helpful. Hard to remember to be comprehensive at 3AM sometimes. >>> >>> Any guidelines for server specs? >>> >>> WILL >>> >>> >>> On Wed, Apr 10, 2013 at 6:10 AM, Lukas Tribus <[email protected]>wrote: >>> >>>> > Unfortunately, I restarted haproxy, so show errors >>>> > returns little of use. >>>> >>>> Alright, since we don't know what actually happened and >>>> there are no similar reports, we will probably not find >>>> the root cause this time. >>>> >>>> Next time this happens, please collect the information >>>> I asked using "show errors", strace -p<pid> and check >>>> the exact cpu usage with top, vmstat and free -m. >>>> >>>> >>>> Regards, >>>> Lukas >>> >>> >>> >>> >>> -- >>> Forio Online Simulations >>> >>> Will Glass-Husain >>> 415 440-7500x89 >>> [email protected] >>> www.forio.com >>> >> >> > > > -- > Forio Online Simulations > > Will Glass-Husain > 415 440-7500x89 > [email protected] > www.forio.com >
