From fa1ac24e1915a410742f047e4f2ce491e25837d1 Mon Sep 17 00:00:00 2001 From: Yingrong Zhao Date: Thu, 24 Oct 2019 12:24:42 -0400 Subject: [PATCH] satellite/gracefulexit: add failure threshold check (#3329) * add overall failure percentage check and inactive time frame check before sending a response to sno * update comment * delete node from transfer queue if it has been inactive for too long * fix linting error * add test config value * fix nil pointer * add config value into testplanet * add unit test for overall failure threshold * move timeframe threshold to chore * update protolock * add chore test * add per peiece failure count logic * change config name from EndpointMaxFailures to MaxFailuresPerPiece * address comments * fix linting error * add error handling for no row returned from progress table * fix test for graceful exit chore on storagenode * fix typo InActive -> Inactive * improve readability for failure threshold calculation * update config lock * change error handling for GetProgress in graceful exit endpoint on the satellite side * return proper rpc error in endpoint * add check in chore test for checking finish timestamp and queue --- internal/testplanet/satellite.go | 6 +- pkg/pb/gracefulexit.pb.go | 151 +++++++++++--------- pkg/pb/gracefulexit.proto | 2 + proto.lock | 8 ++ satellite/gracefulexit/chore.go | 51 ++++++- satellite/gracefulexit/chore_test.go | 50 ++++++- satellite/gracefulexit/common.go | 8 +- satellite/gracefulexit/endpoint.go | 59 ++++++-- satellite/gracefulexit/endpoint_test.go | 16 ++- satellite/gracefulexit/getnodes_test.go | 16 ++- satellite/overlay/service.go | 4 +- satellite/satellitedb/gracefulexit.go | 2 +- satellite/satellitedb/overlaycache.go | 39 +---- scripts/testdata/satellite-config.yaml.lock | 8 +- storagenode/gracefulexit/chore_test.go | 2 +- 15 files changed, 279 insertions(+), 143 deletions(-) diff --git a/internal/testplanet/satellite.go b/internal/testplanet/satellite.go index cf2111c82..613681d52 100644 --- a/internal/testplanet/satellite.go +++ b/internal/testplanet/satellite.go @@ -363,8 +363,10 @@ func (planet *Planet) newSatellites(count int) ([]*SatelliteSystem, error) { ChoreBatchSize: 10, ChoreInterval: defaultInterval, - EndpointBatchSize: 100, - EndpointMaxFailures: 5, + EndpointBatchSize: 100, + MaxFailuresPerPiece: 5, + MaxInactiveTimeFrame: time.Second * 10, + OverallMaxFailuresPercentage: 10, }, Metrics: metrics.Config{ ChoreInterval: defaultInterval, diff --git a/pkg/pb/gracefulexit.pb.go b/pkg/pb/gracefulexit.pb.go index 20bcb13c7..b18e03be8 100644 --- a/pkg/pb/gracefulexit.pb.go +++ b/pkg/pb/gracefulexit.pb.go @@ -58,15 +58,21 @@ func (TransferFailed_Error) EnumDescriptor() ([]byte, []int) { type ExitFailed_Reason int32 const ( - ExitFailed_VERIFICATION_FAILED ExitFailed_Reason = 0 + ExitFailed_VERIFICATION_FAILED ExitFailed_Reason = 0 + ExitFailed_INACTIVE_TIMEFRAME_EXCEEDED ExitFailed_Reason = 1 + ExitFailed_OVERALL_FAILURE_PERCENTAGE_EXCEEDED ExitFailed_Reason = 2 ) var ExitFailed_Reason_name = map[int32]string{ 0: "VERIFICATION_FAILED", + 1: "INACTIVE_TIMEFRAME_EXCEEDED", + 2: "OVERALL_FAILURE_PERCENTAGE_EXCEEDED", } var ExitFailed_Reason_value = map[string]int32{ - "VERIFICATION_FAILED": 0, + "VERIFICATION_FAILED": 0, + "INACTIVE_TIMEFRAME_EXCEEDED": 1, + "OVERALL_FAILURE_PERCENTAGE_EXCEEDED": 2, } func (x ExitFailed_Reason) String() string { @@ -1032,75 +1038,78 @@ func init() { func init() { proto.RegisterFile("gracefulexit.proto", fileDescriptor_8f0acbf2ce5fa631) } var fileDescriptor_8f0acbf2ce5fa631 = []byte{ - // 1076 bytes of a gzipped FileDescriptorProto - 0x1f, 0x8b, 0x08, 0x00, 0x00, 0x00, 0x00, 0x00, 0x02, 0xff, 0xa4, 0x56, 0xef, 0x6e, 0x1b, 0x45, - 0x10, 0xf7, 0xb9, 0x8d, 0x13, 0x8f, 0x1d, 0xc7, 0xd9, 0xfc, 0x33, 0x6e, 0x1b, 0xbb, 0x27, 0x10, - 0xae, 0x04, 0x11, 0x32, 0x50, 0x90, 0x2a, 0x81, 0x9c, 0xda, 0xb1, 0x4d, 0xc3, 0x39, 0x5d, 0x27, - 0x05, 0x21, 0xc1, 0x69, 0xeb, 0x9b, 0x38, 0x27, 0xec, 0xbb, 0x63, 0x6f, 0x5d, 0x35, 0x5f, 0x78, - 0x0e, 0x3e, 0xf1, 0x01, 0x5e, 0x06, 0x1e, 0x01, 0x24, 0x2a, 0xf1, 0x26, 0x68, 0xef, 0xf6, 0x2e, - 0x77, 0xb6, 0x13, 0x22, 0xf5, 0x93, 0x7d, 0x33, 0xbf, 0xf9, 0xed, 0xec, 0xcc, 0x6f, 0x67, 0x17, - 0xc8, 0x98, 0xb3, 0x11, 0x9e, 0xcf, 0x26, 0xf8, 0xda, 0x16, 0x07, 0x1e, 0x77, 0x85, 0x4b, 0x8a, - 0x49, 0x5b, 0x15, 0xc6, 0xee, 0xd8, 0x0d, 0x3d, 0xd5, 0xd2, 0x14, 0x05, 0xb3, 0x9d, 0xf3, 0xe8, - 0xbb, 0xe8, 0x72, 0x0b, 0xb9, 0x1f, 0x7e, 0xe9, 0x75, 0xd8, 0xef, 0xa2, 0x30, 0x5c, 0xa7, 0xf3, - 0xda, 0x16, 0xb6, 0x33, 0x1e, 0x32, 0x81, 0x93, 0x89, 0x2d, 0xd0, 0xa7, 0xf8, 0xd3, 0x0c, 0x7d, - 0xa1, 0x1f, 0xc1, 0xbd, 0xbe, 0x63, 0x0b, 0x9b, 0x09, 0xec, 0xaa, 0x35, 0x24, 0x56, 0xb9, 0xc9, - 0xfb, 0xb0, 0xea, 0xb8, 0x16, 0x9a, 0xb6, 0x55, 0xd1, 0xea, 0x5a, 0xa3, 0x78, 0x58, 0xfa, 0xe3, - 0x4d, 0x2d, 0xf3, 0xf7, 0x9b, 0x5a, 0xce, 0x70, 0x2d, 0xec, 0xb7, 0x69, 0x4e, 0xba, 0xfb, 0x96, - 0xfe, 0x33, 0x6c, 0x2d, 0x59, 0xe6, 0xd6, 0xf1, 0xa4, 0x06, 0x05, 0xcb, 0x9d, 0x32, 0xdb, 0x31, - 0x1d, 0x36, 0xc5, 0x4a, 0xb6, 0xae, 0x35, 0xf2, 0x14, 0x42, 0x93, 0xc1, 0xa6, 0x48, 0x1e, 0x00, - 0xf8, 0x1e, 0x1b, 0xa1, 0x39, 0xf3, 0xd1, 0xaa, 0xdc, 0xa9, 0x6b, 0x0d, 0x8d, 0xe6, 0x03, 0xcb, - 0x99, 0x8f, 0x96, 0x6e, 0x41, 0xed, 0xda, 0x9d, 0xfa, 0x9e, 0xeb, 0xf8, 0x48, 0x5a, 0x00, 0x7e, - 0x6c, 0xad, 0x68, 0xf5, 0x3b, 0x8d, 0x42, 0xf3, 0xe1, 0x41, 0xaa, 0xda, 0x4b, 0xe2, 0x69, 0x22, - 0x48, 0xaf, 0xc0, 0x6e, 0x17, 0x85, 0x84, 0x9c, 0x70, 0x77, 0xcc, 0xd1, 0x8f, 0xeb, 0xf8, 0x1c, - 0xf6, 0x16, 0x3c, 0x6a, 0xdd, 0xc7, 0xb0, 0xe6, 0x29, 0x9b, 0x5a, 0xb5, 0x9a, 0x5e, 0x35, 0x15, - 0x15, 0x63, 0xf5, 0xdf, 0x35, 0x28, 0x26, 0x5d, 0xf3, 0x35, 0xd2, 0x16, 0x6a, 0x94, 0xa8, 0x76, - 0xf6, 0xc6, 0x6a, 0x3f, 0x82, 0xb2, 0x87, 0x7c, 0x84, 0x8e, 0x30, 0x47, 0xee, 0xd4, 0x9b, 0xa0, - 0xc0, 0xa0, 0xa4, 0x59, 0xba, 0xa1, 0xec, 0x4f, 0x95, 0x99, 0xec, 0x03, 0xf8, 0xb3, 0xd1, 0x08, - 0x7d, 0xff, 0x7c, 0x36, 0xa9, 0xdc, 0xad, 0x6b, 0x8d, 0x35, 0x9a, 0xb0, 0xe8, 0xbf, 0x65, 0x61, - 0xf3, 0x94, 0x33, 0xc7, 0x3f, 0x47, 0x3e, 0x94, 0x66, 0xb4, 0xd0, 0x22, 0x6d, 0xd8, 0x76, 0xb9, - 0x3d, 0xb6, 0x1d, 0x36, 0x31, 0x03, 0x45, 0x9a, 0x13, 0x7b, 0x6a, 0x8b, 0x20, 0xe7, 0x42, 0x93, - 0x1c, 0x28, 0x95, 0x0e, 0xe4, 0xcf, 0xb1, 0xf4, 0x50, 0x12, 0xe1, 0xaf, 0x6c, 0xa4, 0x05, 0x5b, - 0x31, 0x8b, 0x67, 0xe3, 0x08, 0xcd, 0x0b, 0xe6, 0x5f, 0x04, 0x7b, 0x2b, 0x34, 0x37, 0x23, 0x92, - 0x13, 0xe9, 0xe9, 0x31, 0xff, 0x82, 0x6e, 0x46, 0xe8, 0xd8, 0x44, 0xba, 0xb0, 0xcb, 0xd1, 0x9b, - 0xb0, 0x11, 0x4e, 0xe5, 0x6e, 0x13, 0x2c, 0x77, 0xae, 0x63, 0xd9, 0x4e, 0x04, 0x5c, 0x11, 0x3d, - 0x81, 0xcd, 0xb9, 0x5c, 0x6c, 0x2b, 0x28, 0x47, 0xf1, 0x70, 0x43, 0x55, 0x79, 0x35, 0x40, 0xf7, - 0xdb, 0x74, 0x23, 0x95, 0x47, 0xdf, 0xd2, 0xff, 0xd5, 0xa0, 0x14, 0x15, 0xe9, 0x88, 0xd9, 0x13, - 0xb4, 0x96, 0xf3, 0x69, 0xb7, 0xe3, 0x23, 0x9f, 0xc3, 0x0a, 0x72, 0xee, 0xf2, 0xa0, 0x14, 0xa5, - 0xa6, 0x9e, 0xd6, 0x53, 0x7a, 0xa5, 0x83, 0x8e, 0x44, 0xd2, 0x30, 0x40, 0xff, 0x16, 0x56, 0x82, - 0x6f, 0xb2, 0x0e, 0x79, 0x63, 0x70, 0x6a, 0x1e, 0x0d, 0xce, 0x8c, 0x76, 0x39, 0x43, 0xee, 0x43, - 0x65, 0x78, 0x3a, 0xa0, 0xad, 0x6e, 0xc7, 0x34, 0x06, 0xed, 0x8e, 0x79, 0x66, 0xb4, 0x5e, 0xb4, - 0xfa, 0xc7, 0xad, 0xc3, 0xe3, 0x4e, 0x59, 0x23, 0x3b, 0xb0, 0xd9, 0x6b, 0x0d, 0x7b, 0xe6, 0x8b, - 0x0e, 0xed, 0x1f, 0xf5, 0x9f, 0xb6, 0x4e, 0xfb, 0x03, 0xa3, 0x9c, 0x25, 0x05, 0x58, 0x3d, 0x33, - 0x9e, 0x19, 0x83, 0x6f, 0x8c, 0x32, 0xe8, 0xbf, 0x68, 0x40, 0x86, 0xc2, 0xe5, 0x6c, 0x8c, 0x52, - 0x6d, 0x5f, 0xa3, 0xef, 0xb3, 0x31, 0x92, 0x2f, 0x21, 0xef, 0x47, 0xb2, 0x50, 0xed, 0xaf, 0x2d, - 0x4f, 0x37, 0x56, 0x4f, 0x2f, 0x43, 0xaf, 0x62, 0xc8, 0x63, 0xc8, 0x9d, 0x07, 0x1b, 0x51, 0x7d, - 0xbf, 0x7f, 0xd3, 0x66, 0x7b, 0x19, 0xaa, 0xd0, 0x87, 0x79, 0x58, 0x55, 0x39, 0xe8, 0x00, 0x6b, - 0x86, 0x2b, 0x28, 0x32, 0xeb, 0x52, 0xff, 0x4b, 0x83, 0xf5, 0x28, 0x26, 0x28, 0xe7, 0xdb, 0x76, - 0xa2, 0xe0, 0x71, 0xfb, 0x15, 0x13, 0x68, 0xfe, 0x88, 0x97, 0xea, 0xd8, 0xed, 0xa9, 0xb0, 0x8d, - 0x00, 0x75, 0x12, 0xfa, 0x9f, 0xe1, 0x25, 0x05, 0x2f, 0xfe, 0x4f, 0x9e, 0xc3, 0x0e, 0xb3, 0x2c, - 0x79, 0xb0, 0xd1, 0x4a, 0x9d, 0x91, 0x50, 0x98, 0x0f, 0x0e, 0xe2, 0xc9, 0xde, 0x8a, 0x60, 0x89, - 0xe3, 0xb2, 0xc5, 0x16, 0x8d, 0xfa, 0x57, 0x50, 0x68, 0xa3, 0x3c, 0xb5, 0x6f, 0xbf, 0x31, 0xbd, - 0x0b, 0xeb, 0x72, 0xf8, 0x44, 0x73, 0x40, 0xf6, 0x61, 0x4f, 0x16, 0x3c, 0x1e, 0x18, 0xa6, 0x6f, - 0x8f, 0x1d, 0x26, 0x66, 0x3c, 0x9c, 0x44, 0x45, 0xba, 0x83, 0x09, 0xfc, 0x30, 0x72, 0xea, 0xbf, - 0x6a, 0x00, 0x92, 0x49, 0xe9, 0xfe, 0x13, 0xd8, 0x0d, 0x68, 0x64, 0x97, 0x66, 0x7c, 0x91, 0x65, - 0x1b, 0x15, 0x76, 0xc6, 0xaf, 0x48, 0xc8, 0x67, 0x90, 0xe3, 0xc8, 0x7c, 0xd7, 0x51, 0x8a, 0xaf, - 0x2d, 0x4e, 0x50, 0xa5, 0x76, 0x1a, 0xc0, 0xa8, 0x82, 0xeb, 0x0f, 0x21, 0x17, 0x5a, 0xc8, 0x1e, - 0x6c, 0x25, 0xe5, 0x6b, 0x1e, 0xb5, 0xfa, 0xc7, 0x9d, 0x76, 0x39, 0xa3, 0xff, 0x93, 0x85, 0x72, - 0x3c, 0xee, 0x23, 0xd9, 0x7e, 0x0a, 0x79, 0xc7, 0x15, 0x26, 0x97, 0x9a, 0x51, 0xb2, 0xdd, 0x9d, - 0xbf, 0x2b, 0x42, 0x45, 0xf5, 0x32, 0x74, 0xcd, 0x51, 0xff, 0x49, 0x1b, 0x4a, 0x42, 0x89, 0x2b, - 0x2c, 0xb9, 0x12, 0xed, 0xbd, 0xe5, 0xa2, 0x0d, 0xc7, 0x4b, 0x86, 0xae, 0x8b, 0x94, 0x22, 0xbf, - 0x80, 0xa2, 0x15, 0xf4, 0x51, 0x71, 0x84, 0x8a, 0x78, 0x27, 0xcd, 0x91, 0xe8, 0x74, 0x2f, 0x43, - 0x0b, 0x56, 0xa2, 0xf1, 0x6d, 0x28, 0xa5, 0x5a, 0x15, 0x0e, 0xaa, 0x85, 0x2c, 0x52, 0xfd, 0x95, - 0x59, 0x60, 0xaa, 0xe1, 0x4f, 0xa0, 0x10, 0x77, 0x0a, 0xad, 0xca, 0x4a, 0x40, 0x51, 0xb9, 0xae, - 0xf0, 0xbd, 0x0c, 0x05, 0x8c, 0xbf, 0x12, 0xa7, 0xaf, 0xf9, 0x67, 0x16, 0xca, 0x72, 0x22, 0x24, - 0xdf, 0x17, 0xe4, 0x55, 0x70, 0x5f, 0x2e, 0xbb, 0xaf, 0xc9, 0x07, 0xe9, 0x25, 0x6e, 0x7e, 0xc0, - 0x54, 0x3f, 0xbc, 0x25, 0x5a, 0x5d, 0xc6, 0xdf, 0xc3, 0xf6, 0xb2, 0xf7, 0x0e, 0x79, 0x94, 0xa6, - 0xb9, 0xe1, 0x4d, 0x54, 0xbd, 0xe1, 0xf6, 0x26, 0x3f, 0xc0, 0xc6, 0xdc, 0x33, 0x80, 0xbc, 0xbb, - 0x90, 0xe0, 0x92, 0xf7, 0x43, 0xf5, 0xbd, 0xff, 0x41, 0x85, 0xe9, 0x37, 0x2f, 0x60, 0x27, 0xde, - 0x54, 0x2a, 0xff, 0x01, 0xac, 0x9e, 0x70, 0x57, 0x5e, 0xca, 0xa4, 0x9e, 0xa6, 0x5a, 0x9c, 0xc9, - 0xd5, 0xfd, 0x39, 0xc4, 0x9c, 0xf8, 0x1b, 0xda, 0x47, 0xda, 0xe1, 0xdd, 0xef, 0xb2, 0xde, 0xcb, - 0x97, 0xb9, 0xe0, 0x1d, 0xf9, 0xf1, 0x7f, 0x01, 0x00, 0x00, 0xff, 0xff, 0x9f, 0x3c, 0xc5, 0xf1, - 0x95, 0x0a, 0x00, 0x00, + // 1128 bytes of a gzipped FileDescriptorProto + 0x1f, 0x8b, 0x08, 0x00, 0x00, 0x00, 0x00, 0x00, 0x02, 0xff, 0xa4, 0x56, 0xef, 0x6e, 0xe3, 0x44, + 0x10, 0x8f, 0x73, 0xd7, 0xb4, 0x99, 0xa4, 0x69, 0xba, 0xfd, 0x17, 0xd2, 0xbb, 0x26, 0x18, 0xd0, + 0xe5, 0x24, 0xa8, 0x50, 0x80, 0x03, 0xe9, 0x24, 0x90, 0xdb, 0x38, 0x89, 0xb9, 0xd4, 0xe9, 0x6d, + 0xd2, 0x72, 0x42, 0x02, 0xcb, 0x17, 0x4f, 0x53, 0xeb, 0x12, 0x3b, 0xac, 0x9d, 0xd3, 0xf5, 0x0b, + 0xcf, 0xc1, 0x67, 0x78, 0x19, 0x78, 0x04, 0x90, 0x38, 0x89, 0x6f, 0x3c, 0x06, 0x5a, 0x7b, 0xe3, + 0xda, 0x49, 0x5a, 0x2a, 0xdd, 0xa7, 0xc4, 0x33, 0xbf, 0x99, 0x9d, 0x9d, 0xf9, 0xcd, 0xec, 0x00, + 0x19, 0x32, 0x73, 0x80, 0x17, 0xd3, 0x11, 0xbe, 0xb1, 0xfd, 0xc3, 0x09, 0x73, 0x7d, 0x97, 0xe4, + 0xe3, 0xb2, 0x32, 0x0c, 0xdd, 0xa1, 0x1b, 0x6a, 0xca, 0x85, 0x31, 0xfa, 0xa6, 0xed, 0x5c, 0xcc, + 0xbe, 0xf3, 0x2e, 0xb3, 0x90, 0x79, 0xe1, 0x97, 0x5c, 0x85, 0x83, 0x16, 0xfa, 0xba, 0xeb, 0xa8, + 0x6f, 0x6c, 0xdf, 0x76, 0x86, 0x3d, 0xd3, 0xc7, 0xd1, 0xc8, 0xf6, 0xd1, 0xa3, 0xf8, 0xd3, 0x14, + 0x3d, 0x5f, 0x6e, 0xc2, 0xbe, 0xe6, 0xd8, 0xbe, 0x6d, 0xfa, 0xd8, 0x12, 0x67, 0x70, 0xac, 0x50, + 0x93, 0x47, 0xb0, 0xea, 0xb8, 0x16, 0x1a, 0xb6, 0x55, 0x92, 0xaa, 0x52, 0x2d, 0x7f, 0x54, 0xf8, + 0xfd, 0x6d, 0x25, 0xf5, 0xd7, 0xdb, 0x4a, 0x46, 0x77, 0x2d, 0xd4, 0x1a, 0x34, 0xc3, 0xd5, 0x9a, + 0x25, 0xff, 0x0c, 0x5b, 0x4b, 0x8e, 0xb9, 0xb3, 0x3d, 0xa9, 0x40, 0xce, 0x72, 0xc7, 0xa6, 0xed, + 0x18, 0x8e, 0x39, 0xc6, 0x52, 0xba, 0x2a, 0xd5, 0xb2, 0x14, 0x42, 0x91, 0x6e, 0x8e, 0x91, 0x3c, + 0x04, 0xf0, 0x26, 0xe6, 0x00, 0x8d, 0xa9, 0x87, 0x56, 0xe9, 0x5e, 0x55, 0xaa, 0x49, 0x34, 0x1b, + 0x48, 0xce, 0x3c, 0xb4, 0x64, 0x0b, 0x2a, 0x37, 0xde, 0xd4, 0x9b, 0xb8, 0x8e, 0x87, 0x44, 0x01, + 0xf0, 0x22, 0x69, 0x49, 0xaa, 0xde, 0xab, 0xe5, 0xea, 0xef, 0x1f, 0x26, 0xb2, 0xbd, 0xc4, 0x9e, + 0xc6, 0x8c, 0xe4, 0x12, 0xec, 0xb6, 0xd0, 0xe7, 0x90, 0x53, 0xe6, 0x0e, 0x19, 0x7a, 0x51, 0x1e, + 0x9f, 0xc3, 0xde, 0x82, 0x46, 0x9c, 0xfb, 0x04, 0xd6, 0x26, 0x42, 0x26, 0x4e, 0x2d, 0x27, 0x4f, + 0x4d, 0x58, 0x45, 0x58, 0xf9, 0x37, 0x09, 0xf2, 0x71, 0xd5, 0x7c, 0x8e, 0xa4, 0x85, 0x1c, 0xc5, + 0xb2, 0x9d, 0xbe, 0x35, 0xdb, 0x8f, 0xa1, 0x38, 0x41, 0x36, 0x40, 0xc7, 0x37, 0x06, 0xee, 0x78, + 0x32, 0x42, 0x1f, 0x83, 0x94, 0xa6, 0xe9, 0x86, 0x90, 0x1f, 0x0b, 0x31, 0x39, 0x00, 0xf0, 0xa6, + 0x83, 0x01, 0x7a, 0xde, 0xc5, 0x74, 0x54, 0xba, 0x5f, 0x95, 0x6a, 0x6b, 0x34, 0x26, 0x91, 0x7f, + 0x4d, 0xc3, 0x66, 0x9f, 0x99, 0x8e, 0x77, 0x81, 0xac, 0xc7, 0xc5, 0x68, 0xa1, 0x45, 0x1a, 0xb0, + 0xed, 0x32, 0x7b, 0x68, 0x3b, 0xe6, 0xc8, 0x08, 0x18, 0x69, 0x8c, 0xec, 0xb1, 0xed, 0x07, 0x31, + 0xe7, 0xea, 0xe4, 0x50, 0xb0, 0xb4, 0xcb, 0x7f, 0x3a, 0x5c, 0x43, 0xc9, 0x0c, 0x7f, 0x2d, 0x23, + 0x0a, 0x6c, 0x45, 0x5e, 0x26, 0x36, 0x0e, 0xd0, 0xb8, 0x34, 0xbd, 0xcb, 0xe0, 0x6e, 0xb9, 0xfa, + 0xe6, 0xcc, 0xc9, 0x29, 0xd7, 0xb4, 0x4d, 0xef, 0x92, 0x6e, 0xce, 0xd0, 0x91, 0x88, 0xb4, 0x60, + 0x97, 0xe1, 0x64, 0x64, 0x0e, 0x70, 0xcc, 0x6f, 0x1b, 0xf3, 0x72, 0xef, 0x26, 0x2f, 0xdb, 0x31, + 0x83, 0x6b, 0x47, 0x4f, 0x61, 0x73, 0x2e, 0x16, 0xdb, 0x0a, 0xd2, 0x91, 0x3f, 0xda, 0x10, 0x59, + 0x5e, 0x0d, 0xd0, 0x5a, 0x83, 0x6e, 0x24, 0xe2, 0xd0, 0x2c, 0xf9, 0x1f, 0x09, 0x0a, 0xb3, 0x24, + 0x35, 0x4d, 0x7b, 0x84, 0xd6, 0x72, 0x7f, 0xd2, 0xdd, 0xfc, 0x91, 0xaf, 0x60, 0x05, 0x19, 0x73, + 0x59, 0x90, 0x8a, 0x42, 0x5d, 0x4e, 0xf2, 0x29, 0x79, 0xd2, 0xa1, 0xca, 0x91, 0x34, 0x34, 0x90, + 0x5f, 0xc0, 0x4a, 0xf0, 0x4d, 0xd6, 0x21, 0xab, 0x77, 0xfb, 0x46, 0xb3, 0x7b, 0xa6, 0x37, 0x8a, + 0x29, 0xf2, 0x00, 0x4a, 0xbd, 0x7e, 0x97, 0x2a, 0x2d, 0xd5, 0xd0, 0xbb, 0x0d, 0xd5, 0x38, 0xd3, + 0x95, 0x73, 0x45, 0xeb, 0x28, 0x47, 0x1d, 0xb5, 0x28, 0x91, 0x1d, 0xd8, 0x6c, 0x2b, 0xbd, 0xb6, + 0x71, 0xae, 0x52, 0xad, 0xa9, 0x1d, 0x2b, 0x7d, 0xad, 0xab, 0x17, 0xd3, 0x24, 0x07, 0xab, 0x67, + 0xfa, 0x33, 0xbd, 0xfb, 0x9d, 0x5e, 0x04, 0xf9, 0x17, 0x09, 0x48, 0xcf, 0x77, 0x99, 0x39, 0x44, + 0xce, 0xb6, 0x13, 0xf4, 0x3c, 0x73, 0x88, 0xe4, 0x1b, 0xc8, 0x7a, 0x33, 0x5a, 0x88, 0xf2, 0x57, + 0x96, 0x87, 0x1b, 0xb1, 0xa7, 0x9d, 0xa2, 0xd7, 0x36, 0xe4, 0x09, 0x64, 0x2e, 0x82, 0x8b, 0x88, + 0xba, 0x3f, 0xb8, 0xed, 0xb2, 0xed, 0x14, 0x15, 0xe8, 0xa3, 0x2c, 0xac, 0x8a, 0x18, 0x64, 0x80, + 0x35, 0xdd, 0xf5, 0x29, 0x9a, 0xd6, 0x95, 0xfc, 0xa7, 0x04, 0xeb, 0x33, 0x9b, 0x20, 0x9d, 0xef, + 0x5a, 0x89, 0xdc, 0x84, 0xd9, 0xaf, 0x4d, 0x1f, 0x8d, 0x57, 0x78, 0x25, 0xda, 0x6e, 0x4f, 0x98, + 0x6d, 0x04, 0xa8, 0xd3, 0x50, 0xff, 0x0c, 0xaf, 0x28, 0x4c, 0xa2, 0xff, 0xe4, 0x39, 0xec, 0x98, + 0x96, 0xc5, 0x1b, 0x1b, 0xad, 0x44, 0x8f, 0x84, 0xc4, 0x7c, 0x78, 0x18, 0x4d, 0x76, 0x65, 0x06, + 0x8b, 0xb5, 0xcb, 0x96, 0xb9, 0x28, 0x94, 0xbf, 0x85, 0x5c, 0x03, 0x79, 0xd7, 0xbe, 0xfb, 0xc5, + 0xe4, 0x16, 0xac, 0xf3, 0xe1, 0x33, 0x9b, 0x03, 0xbc, 0x0e, 0x7b, 0x3c, 0xe1, 0xd1, 0xc0, 0x30, + 0x3c, 0x7b, 0xe8, 0x98, 0xfe, 0x94, 0x85, 0x93, 0x28, 0x4f, 0x77, 0x30, 0x86, 0xef, 0xcd, 0x94, + 0xf2, 0xbf, 0x12, 0x00, 0xf7, 0x24, 0x78, 0xff, 0x39, 0xec, 0x06, 0x6e, 0x78, 0x95, 0xa6, 0x6c, + 0xd1, 0xcb, 0x36, 0x0a, 0xec, 0x94, 0x5d, 0x3b, 0x21, 0x5f, 0x42, 0x86, 0xa1, 0xe9, 0xb9, 0x8e, + 0x60, 0x7c, 0x65, 0x71, 0x82, 0x0a, 0xb6, 0xd3, 0x00, 0x46, 0x05, 0x5c, 0x7e, 0x05, 0x99, 0x50, + 0x42, 0xf6, 0x60, 0x2b, 0x4e, 0x5f, 0xa3, 0xa9, 0x68, 0x1d, 0x95, 0x53, 0xbf, 0x02, 0xfb, 0x9a, + 0xae, 0x1c, 0xf7, 0xb5, 0x73, 0xd5, 0xe8, 0x6b, 0x27, 0x6a, 0x93, 0x2a, 0x27, 0xaa, 0xa1, 0xbe, + 0x38, 0x56, 0xd5, 0x86, 0xda, 0x28, 0x4a, 0xe4, 0x11, 0x7c, 0xd0, 0x3d, 0x57, 0xa9, 0xd2, 0xe9, + 0x04, 0x46, 0x67, 0x54, 0x35, 0x4e, 0x55, 0x7a, 0xac, 0xea, 0x7d, 0xde, 0x2e, 0x11, 0x30, 0x2d, + 0xff, 0x9d, 0x86, 0x62, 0xf4, 0x70, 0xcc, 0x1a, 0xe0, 0x0b, 0xc8, 0x3a, 0xae, 0x6f, 0x30, 0xce, + 0x3e, 0xd1, 0x00, 0xbb, 0xf3, 0xaf, 0x4e, 0xc8, 0xcd, 0x76, 0x8a, 0xae, 0x39, 0xe2, 0x3f, 0x69, + 0x40, 0xc1, 0x17, 0x34, 0x0d, 0x8b, 0x27, 0xe8, 0xbf, 0xbf, 0x9c, 0xfe, 0xe1, 0xa0, 0x4a, 0xd1, + 0x75, 0x3f, 0xc1, 0xed, 0xaf, 0x21, 0x6f, 0x05, 0x8c, 0x10, 0x3e, 0x42, 0x6e, 0xbd, 0x97, 0xf4, + 0x11, 0xe3, 0x4c, 0x3b, 0x45, 0x73, 0x56, 0x8c, 0x42, 0x0d, 0x28, 0x24, 0x8a, 0x1e, 0x8e, 0xbc, + 0x85, 0x28, 0x12, 0x4c, 0xe1, 0x51, 0x60, 0x82, 0x3a, 0x4f, 0x21, 0x17, 0xd5, 0x1c, 0xad, 0xd2, + 0x4a, 0xe0, 0xa2, 0x74, 0x53, 0x09, 0xdb, 0x29, 0x0a, 0x18, 0x7d, 0xc5, 0xfa, 0xb8, 0xfe, 0x47, + 0x1a, 0x8a, 0x7c, 0xb6, 0xc4, 0x37, 0x15, 0xf2, 0x3a, 0x78, 0x79, 0x97, 0xbd, 0xfc, 0xe4, 0xe3, + 0xe4, 0x11, 0xb7, 0xaf, 0x42, 0xe5, 0x4f, 0xee, 0x88, 0x16, 0xcf, 0xfa, 0x0f, 0xb0, 0xbd, 0x6c, + 0x73, 0x22, 0x8f, 0x93, 0x6e, 0x6e, 0xd9, 0xae, 0xca, 0xb7, 0xec, 0x01, 0xe4, 0x47, 0xd8, 0x98, + 0x5b, 0x28, 0xc8, 0x87, 0x0b, 0x01, 0x2e, 0xd9, 0x44, 0xca, 0x1f, 0xfd, 0x0f, 0x2a, 0x0c, 0xbf, + 0x7e, 0x09, 0x3b, 0xd1, 0xa5, 0x12, 0xf1, 0x77, 0x61, 0xf5, 0x94, 0xb9, 0xfc, 0x79, 0x27, 0xd5, + 0xa4, 0xab, 0xc5, 0xe9, 0x5e, 0x3e, 0x98, 0x43, 0xcc, 0x91, 0xbf, 0x26, 0x7d, 0x2a, 0x1d, 0xdd, + 0xff, 0x3e, 0x3d, 0x79, 0xf9, 0x32, 0x13, 0x6c, 0xa4, 0x9f, 0xfd, 0x17, 0x00, 0x00, 0xff, 0xff, + 0x42, 0x7c, 0x67, 0xa5, 0xdf, 0x0a, 0x00, 0x00, } type DRPCNodeGracefulExitClient interface { diff --git a/pkg/pb/gracefulexit.proto b/pkg/pb/gracefulexit.proto index 4f31d5a4b..e7106059e 100644 --- a/pkg/pb/gracefulexit.proto +++ b/pkg/pb/gracefulexit.proto @@ -103,6 +103,8 @@ message ExitCompleted { message ExitFailed { enum Reason { VERIFICATION_FAILED = 0; + INACTIVE_TIMEFRAME_EXCEEDED = 1; + OVERALL_FAILURE_PERCENTAGE_EXCEEDED = 2; } // on failure bytes exit_failure_signature = 1; diff --git a/proto.lock b/proto.lock index 15b05ba1f..afa8bbba1 100644 --- a/proto.lock +++ b/proto.lock @@ -876,6 +876,14 @@ "enum_fields": [ { "name": "VERIFICATION_FAILED" + }, + { + "name": "INACTIVE_TIMEFRAME_EXCEEDED", + "integer": 1 + }, + { + "name": "OVERALL_FAILURE_PERCENTAGE_EXCEEDED", + "integer": 2 } ] } diff --git a/satellite/gracefulexit/chore.go b/satellite/gracefulexit/chore.go index 750d5db5a..de3a1fe26 100644 --- a/satellite/gracefulexit/chore.go +++ b/satellite/gracefulexit/chore.go @@ -5,11 +5,14 @@ package gracefulexit import ( "context" + "database/sql" "time" + "github.com/zeebo/errs" "go.uber.org/zap" "storj.io/storj/internal/sync2" + "storj.io/storj/pkg/storj" "storj.io/storj/satellite/metainfo" "storj.io/storj/satellite/overlay" ) @@ -46,9 +49,9 @@ func (chore *Chore) Run(ctx context.Context) (err error) { chore.log.Info("running graceful exit chore.") - exitingNodes, err := chore.overlay.GetExitingNodesLoopIncomplete(ctx) + exitingNodes, err := chore.overlay.GetExitingNodes(ctx) if err != nil { - chore.log.Error("error retrieving nodes that have not completed the metainfo loop.", zap.Error(err)) + chore.log.Error("error retrieving nodes that have not finished exiting", zap.Error(err)) return nil } @@ -58,7 +61,47 @@ func (chore *Chore) Run(ctx context.Context) (err error) { return nil } - pathCollector := NewPathCollector(chore.db, exitingNodes, chore.log, chore.config.ChoreBatchSize) + exitingNodesLoopIncomplete := make(storj.NodeIDList, 0, nodeCount) + for _, node := range exitingNodes { + if node.ExitLoopCompletedAt == nil { + exitingNodesLoopIncomplete = append(exitingNodesLoopIncomplete, node.NodeID) + continue + } + + progress, err := chore.db.GetProgress(ctx, node.NodeID) + if err != nil && !errs.Is(err, sql.ErrNoRows) { + chore.log.Error("error retrieving progress for node", zap.Stringer("Node ID", node.NodeID), zap.Error(err)) + continue + } + + lastActivityTime := *node.ExitLoopCompletedAt + if progress != nil { + lastActivityTime = progress.UpdatedAt + } + + // check inactive timeframe + if lastActivityTime.Add(chore.config.MaxInactiveTimeFrame).Before(time.Now().UTC()) { + exitStatusRequest := &overlay.ExitStatusRequest{ + NodeID: node.NodeID, + ExitSuccess: false, + ExitFinishedAt: time.Now().UTC(), + } + _, err = chore.overlay.UpdateExitStatus(ctx, exitStatusRequest) + if err != nil { + chore.log.Error("error updating exit status", zap.Error(err)) + continue + } + + // remove all items from the transfer queue + err := chore.db.DeleteTransferQueueItems(ctx, node.NodeID) + if err != nil { + chore.log.Error("error deleting node from transfer queue", zap.Error(err)) + } + } + } + + // Populate transfer queue for nodes that have not completed the exit loop yet + pathCollector := NewPathCollector(chore.db, exitingNodesLoopIncomplete, chore.log, chore.config.ChoreBatchSize) err = chore.metainfoLoop.Join(ctx, pathCollector) if err != nil { chore.log.Error("error joining metainfo loop.", zap.Error(err)) @@ -72,7 +115,7 @@ func (chore *Chore) Run(ctx context.Context) (err error) { } now := time.Now().UTC() - for _, nodeID := range exitingNodes { + for _, nodeID := range exitingNodesLoopIncomplete { exitStatus := overlay.ExitStatusRequest{ NodeID: nodeID, ExitLoopCompletedAt: now, diff --git a/satellite/gracefulexit/chore_test.go b/satellite/gracefulexit/chore_test.go index c05bbf106..68d8c7dec 100644 --- a/satellite/gracefulexit/chore_test.go +++ b/satellite/gracefulexit/chore_test.go @@ -9,11 +9,13 @@ import ( "time" "github.com/stretchr/testify/require" + "go.uber.org/zap" "storj.io/storj/internal/memory" "storj.io/storj/internal/testcontext" "storj.io/storj/internal/testplanet" "storj.io/storj/internal/testrand" + "storj.io/storj/pkg/storj" "storj.io/storj/satellite" "storj.io/storj/satellite/gracefulexit" "storj.io/storj/satellite/overlay" @@ -22,10 +24,16 @@ import ( ) func TestChore(t *testing.T) { + var maximumInactiveTimeFrame = time.Second * 1 testplanet.Run(t, testplanet.Config{ SatelliteCount: 1, StorageNodeCount: 8, UplinkCount: 1, + Reconfigure: testplanet.Reconfigure{ + Satellite: func(log *zap.Logger, index int, config *satellite.Config) { + config.GracefulExit.MaxInactiveTimeFrame = maximumInactiveTimeFrame + }, + }, }, func(t *testing.T, ctx *testcontext.Context, planet *testplanet.Planet) { uplinkPeer := planet.Uplinks[0] satellite := planet.Satellites[0] @@ -46,16 +54,22 @@ func TestChore(t *testing.T) { err = uplinkPeer.UploadWithConfig(ctx, satellite, rs, "testbucket", "test/path2", testrand.Bytes(5*memory.KiB)) require.NoError(t, err) - exitStatus := overlay.ExitStatusRequest{ + exitStatusRequest := overlay.ExitStatusRequest{ NodeID: exitingNode.ID(), ExitInitiatedAt: time.Now().UTC(), } - _, err = satellite.Overlay.DB.UpdateExitStatus(ctx, &exitStatus) + _, err = satellite.Overlay.DB.UpdateExitStatus(ctx, &exitStatusRequest) require.NoError(t, err) - nodeIDs, err := satellite.Overlay.DB.GetExitingNodesLoopIncomplete(ctx) + exitingNodes, err := satellite.Overlay.DB.GetExitingNodes(ctx) require.NoError(t, err) + nodeIDs := make(storj.NodeIDList, 0, len(exitingNodes)) + for _, exitingNode := range exitingNodes { + if exitingNode.ExitLoopCompletedAt == nil { + nodeIDs = append(nodeIDs, exitingNode.NodeID) + } + } require.Len(t, nodeIDs, 1) satellite.GracefulExit.Chore.Loop.TriggerWait() @@ -77,9 +91,37 @@ func TestChore(t *testing.T) { require.Len(t, incompleteTransfers, 0) } - nodeIDs, err = satellite.Overlay.DB.GetExitingNodesLoopIncomplete(ctx) + exitingNodes, err = satellite.Overlay.DB.GetExitingNodes(ctx) require.NoError(t, err) + nodeIDs = make(storj.NodeIDList, 0, len(exitingNodes)) + for _, exitingNode := range exitingNodes { + if exitingNode.ExitLoopCompletedAt == nil { + nodeIDs = append(nodeIDs, exitingNode.NodeID) + } + } require.Len(t, nodeIDs, 0) + + satellite.GracefulExit.Chore.Loop.Pause() + err = satellite.DB.GracefulExit().IncrementProgress(ctx, exitingNode.ID(), 0, 0, 0) + require.NoError(t, err) + + incompleteTransfers, err = satellite.DB.GracefulExit().GetIncomplete(ctx, exitingNode.ID(), 20, 0) + require.NoError(t, err) + require.Len(t, incompleteTransfers, 2) + + // node should fail graceful exit if it has been inactive for maximum inactive time frame since last activity + time.Sleep(maximumInactiveTimeFrame + time.Second*1) + satellite.GracefulExit.Chore.Loop.TriggerWait() + + exitStatus, err := satellite.Overlay.DB.GetExitStatus(ctx, exitingNode.ID()) + require.NoError(t, err) + require.False(t, exitStatus.ExitSuccess) + require.NotNil(t, exitStatus.ExitFinishedAt) + + incompleteTransfers, err = satellite.DB.GracefulExit().GetIncomplete(ctx, exitingNode.ID(), 20, 0) + require.NoError(t, err) + require.Len(t, incompleteTransfers, 0) + }) } diff --git a/satellite/gracefulexit/common.go b/satellite/gracefulexit/common.go index 97e7cf943..886f9f9d3 100644 --- a/satellite/gracefulexit/common.go +++ b/satellite/gracefulexit/common.go @@ -25,6 +25,10 @@ type Config struct { ChoreBatchSize int `help:"size of the buffer used to batch inserts into the transfer queue." default:"500"` ChoreInterval time.Duration `help:"how often to run the transfer queue chore." releaseDefault:"30s" devDefault:"10s"` - EndpointBatchSize int `help:"size of the buffer used to batch transfer queue reads and sends to the storage node." default:"100"` - EndpointMaxFailures int `help:"maximum number of transfer failures per piece." default:"3"` + EndpointBatchSize int `help:"size of the buffer used to batch transfer queue reads and sends to the storage node." default:"100"` + + MaxFailuresPerPiece int `help:"maximum number of transfer failures per piece." default:"3"` + // TODO: what's the default number? + OverallMaxFailuresPercentage int `help:"maximum percentage of transfer failures per node." default:"10"` + MaxInactiveTimeFrame time.Duration `help:"maximum inactive time frame of transfer activities per node." default:"500h"` } diff --git a/satellite/gracefulexit/endpoint.go b/satellite/gracefulexit/endpoint.go index e3f01520f..685cad83c 100644 --- a/satellite/gracefulexit/endpoint.go +++ b/satellite/gracefulexit/endpoint.go @@ -202,7 +202,7 @@ func (endpoint *Endpoint) doProcess(stream processStream) (err error) { } if len(incomplete) == 0 { - incomplete, err = endpoint.db.GetIncompleteFailed(ctx, nodeID, endpoint.config.EndpointMaxFailures, endpoint.config.EndpointBatchSize, 0) + incomplete, err = endpoint.db.GetIncompleteFailed(ctx, nodeID, endpoint.config.MaxFailuresPerPiece, endpoint.config.EndpointBatchSize, 0) if err != nil { return handleError(err) } @@ -233,19 +233,58 @@ func (endpoint *Endpoint) doProcess(stream processStream) (err error) { } pendingCount := pending.length() + // if there are no more transfers and the pending queue is empty, send complete if atomic.LoadInt32(&morePiecesFlag) == 0 && pendingCount == 0 { - // TODO check whether failure threshold is met before sending completed - // TODO needs exit signature - transferMsg := &pb.SatelliteMessage{ - Message: &pb.SatelliteMessage_ExitCompleted{ - ExitCompleted: &pb.ExitCompleted{}, - }, + exitStatusRequest := &overlay.ExitStatusRequest{ + NodeID: nodeID, + ExitFinishedAt: time.Now().UTC(), } + + progress, err := endpoint.db.GetProgress(ctx, nodeID) + if err != nil { + return rpcstatus.Error(rpcstatus.Internal, err.Error()) + } + + var transferMsg *pb.SatelliteMessage + processed := progress.PiecesFailed + progress.PiecesTransferred + // check node's exiting progress to see if it has failed passed max failure threshold + if processed > 0 && float64(progress.PiecesFailed)/float64(processed)*100 >= float64(endpoint.config.OverallMaxFailuresPercentage) { + + exitStatusRequest.ExitSuccess = false + // TODO needs signature + transferMsg = &pb.SatelliteMessage{ + Message: &pb.SatelliteMessage_ExitFailed{ + ExitFailed: &pb.ExitFailed{ + Reason: pb.ExitFailed_OVERALL_FAILURE_PERCENTAGE_EXCEEDED, + }, + }, + } + } else { + exitStatusRequest.ExitSuccess = true + // TODO needs signature + transferMsg = &pb.SatelliteMessage{ + Message: &pb.SatelliteMessage_ExitCompleted{ + ExitCompleted: &pb.ExitCompleted{}, + }, + } + } + + _, err = endpoint.overlaydb.UpdateExitStatus(ctx, exitStatusRequest) + if err != nil { + return rpcstatus.Error(rpcstatus.Internal, err.Error()) + } + err = stream.Send(transferMsg) if err != nil { return Error.Wrap(err) } + + // remove remaining items from the queue after notifying nodes about their exit status + err = endpoint.db.DeleteTransferQueueItems(ctx, nodeID) + if err != nil { + return rpcstatus.Error(rpcstatus.Internal, err.Error()) + } break } // skip if there are none pending @@ -425,7 +464,7 @@ func (endpoint *Endpoint) handleSucceeded(ctx context.Context, pending *pendingM } var failed int64 - if transferQueueItem.FailedCount != nil && *transferQueueItem.FailedCount > 0 { + if transferQueueItem.FailedCount != nil && *transferQueueItem.FailedCount >= endpoint.config.MaxFailuresPerPiece { failed = -1 } @@ -476,8 +515,8 @@ func (endpoint *Endpoint) handleFailed(ctx context.Context, pending *pendingMap, return Error.Wrap(err) } - // only increment failed if it hasn't failed before - if failedCount == 1 { + // only increment overall failed count if piece failures has reached the threshold + if failedCount == endpoint.config.MaxFailuresPerPiece { err = endpoint.db.IncrementProgress(ctx, nodeID, 0, 0, 1) if err != nil { return Error.Wrap(err) diff --git a/satellite/gracefulexit/endpoint_test.go b/satellite/gracefulexit/endpoint_test.go index 1c4cecb3e..d33820466 100644 --- a/satellite/gracefulexit/endpoint_test.go +++ b/satellite/gracefulexit/endpoint_test.go @@ -123,6 +123,12 @@ func TestFailure(t *testing.T) { case *pb.SatelliteMessage_ExitCompleted: // TODO test completed signature stuff break + case *pb.SatelliteMessage_ExitFailed: + status, err := satellite.DB.OverlayCache().GetExitStatus(ctx, exitingNode.ID()) + require.NoError(t, err) + require.False(t, status.ExitSuccess) + require.Equal(t, m.ExitFailed.Reason, pb.ExitFailed_OVERALL_FAILURE_PERCENTAGE_EXCEEDED) + break default: t.FailNow() } @@ -161,9 +167,9 @@ func testTransfers(t *testing.T, objects int, verifier func(ctx *testcontext.Con require.NoError(t, err) } // check that there are no exiting nodes. - exitingNodeIDs, err := satellite.DB.OverlayCache().GetExitingNodes(ctx) + exitingNodes, err := satellite.DB.OverlayCache().GetExitingNodes(ctx) require.NoError(t, err) - require.Len(t, exitingNodeIDs, 0) + require.Len(t, exitingNodes, 0) exitingNode, err := findNodeToExit(ctx, planet, objects) require.NoError(t, err) @@ -187,11 +193,11 @@ func testTransfers(t *testing.T, objects int, verifier func(ctx *testcontext.Con switch response.GetMessage().(type) { case *pb.SatelliteMessage_NotReady: // now check that the exiting node is initiated. - exitingNodeIDs, err := satellite.DB.OverlayCache().GetExitingNodes(ctx) + exitingNodes, err := satellite.DB.OverlayCache().GetExitingNodes(ctx) require.NoError(t, err) - require.Len(t, exitingNodeIDs, 1) + require.Len(t, exitingNodes, 1) - require.Equal(t, exitingNode.ID(), exitingNodeIDs[0]) + require.Equal(t, exitingNode.ID(), exitingNodes[0].NodeID) default: t.FailNow() } diff --git a/satellite/gracefulexit/getnodes_test.go b/satellite/gracefulexit/getnodes_test.go index 51563e06e..fdea5f285 100644 --- a/satellite/gracefulexit/getnodes_test.go +++ b/satellite/gracefulexit/getnodes_test.go @@ -70,14 +70,20 @@ func TestGetExitingNodes(t *testing.T) { nodes, err := cache.GetExitingNodes(ctx) require.NoError(t, err) require.Len(t, nodes, exitingCount) - for _, id := range nodes { - require.True(t, exiting[id]) + for _, node := range nodes { + require.True(t, exiting[node.NodeID]) } - nodes, err = cache.GetExitingNodesLoopIncomplete(ctx) + nodes, err = cache.GetExitingNodes(ctx) require.NoError(t, err) - require.Len(t, nodes, exitingLoopIncompleteCount) - for _, id := range nodes { + exitingNodesLoopIncomplete := make(storj.NodeIDList, 0, len(nodes)) + for _, node := range nodes { + if node.ExitLoopCompletedAt == nil { + exitingNodesLoopIncomplete = append(exitingNodesLoopIncomplete, node.NodeID) + } + } + require.Len(t, exitingNodesLoopIncomplete, exitingLoopIncompleteCount) + for _, id := range exitingNodesLoopIncomplete { require.True(t, exitingLoopIncomplete[id]) } }) diff --git a/satellite/overlay/service.go b/satellite/overlay/service.go index 0d8c429c4..023af2e37 100644 --- a/satellite/overlay/service.go +++ b/satellite/overlay/service.go @@ -74,9 +74,7 @@ type DB interface { // UpdateExitStatus is used to update a node's graceful exit status. UpdateExitStatus(ctx context.Context, request *ExitStatusRequest) (stats *NodeStats, err error) // GetExitingNodes returns nodes who have initiated a graceful exit, but have not completed it. - GetExitingNodes(ctx context.Context) (exitingNodes storj.NodeIDList, err error) - // GetExitingNodesLoopIncomplete returns exiting nodes who haven't completed the metainfo loop iteration. - GetExitingNodesLoopIncomplete(ctx context.Context) (exitingNodes storj.NodeIDList, err error) + GetExitingNodes(ctx context.Context) (exitingNodes []*ExitStatus, err error) // GetGracefulExitCompletedByTimeFrame returns nodes who have completed graceful exit within a time window (time window is around graceful exit completion). GetGracefulExitCompletedByTimeFrame(ctx context.Context, begin, end time.Time) (exitedNodes storj.NodeIDList, err error) // GetGracefulExitIncompleteByTimeFrame returns nodes who have initiated, but not completed graceful exit within a time window (time window is around graceful exit initiation). diff --git a/satellite/satellitedb/gracefulexit.go b/satellite/satellitedb/gracefulexit.go index 09410c444..b8090c71c 100644 --- a/satellite/satellitedb/gracefulexit.go +++ b/satellite/satellitedb/gracefulexit.go @@ -224,7 +224,7 @@ func (db *gracefulexitDB) GetIncompleteFailed(ctx context.Context, nodeID storj. WHERE node_id = ? AND finished_at is NULL AND last_failed_at is not NULL - AND failed_count <= ? + AND failed_count < ? ORDER BY durability_ratio asc, queued_at asc LIMIT ? OFFSET ?` rows, err := db.db.Query(db.db.Rebind(sql), nodeID.Bytes(), maxFailures, limit, offset) if err != nil { diff --git a/satellite/satellitedb/overlaycache.go b/satellite/satellitedb/overlaycache.go index 8a72bf25b..a6999277c 100644 --- a/satellite/satellitedb/overlaycache.go +++ b/satellite/satellitedb/overlaycache.go @@ -829,11 +829,11 @@ func (cache *overlaycache) UpdatePieceCounts(ctx context.Context, pieceCounts ma } // GetExitingNodes returns nodes who have initiated a graceful exit, but have not completed it. -func (cache *overlaycache) GetExitingNodes(ctx context.Context) (exitingNodes storj.NodeIDList, err error) { +func (cache *overlaycache) GetExitingNodes(ctx context.Context) (exitingNodes []*overlay.ExitStatus, err error) { defer mon.Task()(&ctx)(&err) rows, err := cache.db.Query(cache.db.Rebind(` - SELECT id FROM nodes + SELECT id, exit_initiated_at, exit_loop_completed_at, exit_finished_at, exit_success FROM nodes WHERE exit_initiated_at IS NOT NULL AND exit_finished_at IS NULL `), @@ -846,41 +846,12 @@ func (cache *overlaycache) GetExitingNodes(ctx context.Context) (exitingNodes st }() for rows.Next() { - var id storj.NodeID - err = rows.Scan(&id) + var exitingNodeStatus overlay.ExitStatus + err = rows.Scan(&exitingNodeStatus.NodeID, &exitingNodeStatus.ExitInitiatedAt, &exitingNodeStatus.ExitLoopCompletedAt, &exitingNodeStatus.ExitFinishedAt, &exitingNodeStatus.ExitSuccess) if err != nil { return nil, err } - exitingNodes = append(exitingNodes, id) - } - return exitingNodes, nil -} - -// GetExitingNodesLoopIncomplete returns exiting nodes who haven't completed the metainfo loop iteration. -func (cache *overlaycache) GetExitingNodesLoopIncomplete(ctx context.Context) (exitingNodes storj.NodeIDList, err error) { - defer mon.Task()(&ctx)(&err) - - rows, err := cache.db.Query(cache.db.Rebind(` - SELECT id FROM nodes - WHERE exit_initiated_at IS NOT NULL - AND exit_loop_completed_at IS NULL - AND exit_finished_at IS NULL - `), - ) - if err != nil { - return nil, err - } - defer func() { - err = errs.Combine(err, rows.Close()) - }() - - for rows.Next() { - var id storj.NodeID - err = rows.Scan(&id) - if err != nil { - return nil, err - } - exitingNodes = append(exitingNodes, id) + exitingNodes = append(exitingNodes, &exitingNodeStatus) } return exitingNodes, nil } diff --git a/scripts/testdata/satellite-config.yaml.lock b/scripts/testdata/satellite-config.yaml.lock index 81c9de197..3f2e42824 100644 --- a/scripts/testdata/satellite-config.yaml.lock +++ b/scripts/testdata/satellite-config.yaml.lock @@ -119,7 +119,13 @@ contact.external-address: "" # graceful-exit.endpoint-batch-size: 100 # maximum number of transfer failures per piece. -# graceful-exit.endpoint-max-failures: 3 +# graceful-exit.max-failures-per-piece: 3 + +# maximum inactive time frame of transfer activities per node. +# graceful-exit.max-inactive-time-frame: 500h0m0s + +# maximum percentage of transfer failures per node. +# graceful-exit.overall-max-failures-percentage: 10 # path to the certificate chain for this identity identity.cert-path: /root/.local/share/storj/identity/satellite/identity.cert diff --git a/storagenode/gracefulexit/chore_test.go b/storagenode/gracefulexit/chore_test.go index 612366b27..da7ed2f8c 100644 --- a/storagenode/gracefulexit/chore_test.go +++ b/storagenode/gracefulexit/chore_test.go @@ -106,7 +106,7 @@ func exitSatellite(ctx context.Context, t *testing.T, planet *testplanet.Planet, exitingNodes, err := satellite1.DB.OverlayCache().GetExitingNodes(ctx) require.NoError(t, err) require.Len(t, exitingNodes, 1) - require.Equal(t, exitingNode.ID(), exitingNodes[0]) + require.Equal(t, exitingNode.ID(), exitingNodes[0].NodeID) queueItems, err := satellite1.DB.GracefulExit().GetIncomplete(ctx, exitStatus.NodeID, 10, 0) require.NoError(t, err)