Rarely do we face Linux kernel hangs on our servers. Recently, two identical servers experienced crashes or hangs and upon power cycling, they began spewing endless error dumps. The servers were eventually restored by turning off power, letting them sit, then powering them back on. This highlights that simply power cycling may not be enough to recover a system in some cases. It also raises questions about the efficacy of power cycles versus cool down periods. The author acknowledges uncertainties about the mechanism involved but concludes that cooling down the servers proved faster and more effective in this scenario.
https://utcc.utoronto.ca/~cks/space/blog/tech/ServerWhenPowerCycleNotEnough