Techniques involving flaky path detection

摘要

A technique oversees a path between a multipathing driver of a host computer and a volume of a data storage array. The technique involves, while the multipathing driver of the host computer sends input/output requests (IOs) to the volume of the data storage array on the path, generating an IOs-Over-Period metric based on outcomes of the IOs, the IOs-Over-Period metric providing a measure of IOs per failure over a period of path operation. The technique further involves performing a comparison operation which compares the IOs-Over-Period metric to a predefined flaky path range having a predefined lower limit and a predefined upper limit. The technique further involves, after performing the comparison operation, outputting a detection signal indicating that the path is (i) flaky when the IOs-Over-Period metric falls within the predefined flaky path range and (ii) non-flaky when the IOs-Over-Period metric falls outside of the predefined flaky path range.

主权项

1. A method of overseeing a path between a multipathing driver of a host computer and a volume of a data storage array, the method comprising:
while the multipathing driver of the host computer sends input/output requests (IOs) to the volume of the data storage array on the path, generating an IOs-Over-Period metric based on outcomes of the IOs, the IOs-Over-Period metric providing a measure of IOs per failure over a period of path operation; performing a comparison operation which compares the IOs-Over-Period metric to a predefined flaky path range having a predefined lower limit and a predefined upper limit; and after performing the comparison operation, outputting a detection signal indicating that the path is (i) flaky when the IOs-Over-Period metric falls within the predefined flaky path range and (ii) non-flaky when the IOs-Over-Period metric falls outside of the predefined flaky path range; wherein the IOs sent from the multipathing driver to the volume on the path include a series of small computer system interface (SCSI) commands; and wherein generating the IOs-Over-Period metric based on the outcomes of the IOs includes assigning a series of monotonically increasing instance numbers to the series of SCSI commands to distinctively identify each SCSI command of the series of SCSI commands with an instance number.