确定性 ISSU 资格.pdf

1、Ryan GarthwaiteDeterministic ISSU QualificationDeterministic ISSU QualificationRyan GarthwaiteNETWORKINGIn-Service Software Upgrade(ISSU)ISSU lets network switches upgrade software without disrupting traffic.If an error occurs,any interruption is minimised.Also referred to as Hitless Upgrade and War

2、m Upgrade/bootWhat is ISSU and ISSU Qualification?ISSU QualificationThe warm upgrade is assessed against a series of criteria and metrics.It must pass all of them to consider this upgrade successful.Some implementations use hardware workaroundsDual supervisorsIncreased hardware costs and complexityB

3、andwidth reductionReduced bandwidth while half the system upgrades at a timeIn SONiC this is fully software basedFor full implementation details refer to How is SONiC ISSU Implemented?OCPREG19-SONiC Reliability,Manageability and ExtensibilityChallenges with ISSU QualificationLink-Aggregation-Control

4、(LACP)Enforcing 150s downtime,no LAG flapsBGPEnforcing BGP restarts 240sDataplaneIntegrating with Control Plane Assistant(ARP/ND)Sub-second ASIC reset,Downtime Need to pre-filter out these expected inconsistenciesMany keys are volatile e.g.temperature,CPU,etc-need to exclude these tooOf the differen

5、cesCommon to see warm/cold has a value set but the opposite doesnt have a value setDifferent representations of the same state“N/A”vs(absent variable)Consistency Between Software Cold vs Warm Boot#WB Values#CB Values#WB excl Volatile#CB excl Volatile#Differing Values4724520423612778574APPL_DBMostly

6、a matchSimilar challenges to STATEUnset default valuesConsistency Between Software Cold vs Warm Boot#WB Values#CB Values#WB excl Volatile#CB excl Volatile#Differing Values521145211652081520836Observations from ProductionSW/SW Inconsistency50%SW/HW Inconsistenc