tendermint/consensus/common_test.go

504 lines
15 KiB
Go
Raw Normal View History

2015-12-01 20:12:01 -08:00
package consensus
import (
"bytes"
new pubsub package comment out failing consensus tests for now rewrite rpc httpclient to use new pubsub package import pubsub as tmpubsub, query as tmquery make event IDs constants EventKey -> EventTypeKey rename EventsPubsub to PubSub mempool does not use pubsub rename eventsSub to pubsub new subscribe API fix channel size issues and consensus tests bugs refactor rpc client add missing discardFromChan method add mutex rename pubsub to eventBus remove IsRunning from WSRPCConnection interface (not needed) add a comment in broadcastNewRoundStepsAndVotes rename registerEventCallbacks to broadcastNewRoundStepsAndVotes See https://dave.cheney.net/2014/03/19/channel-axioms stop eventBuses after reactor tests remove unnecessary Unsubscribe return subscribe helper function move discardFromChan to where it is used subscribe now returns an err this gives us ability to refuse to subscribe if pubsub is at its max capacity. use context for control overflow cache queries handle err when subscribing in replay_test rename testClientID to testSubscriber extract var set channel buffer capacity to 1 in replay_file fix byzantine_test unsubscribe from single event, not all events refactor httpclient to return events to appropriate channels return failing testReplayCrashBeforeWriteVote test fix TestValidatorSetChanges refactor code a bit fix testReplayCrashBeforeWriteVote add comment fix TestValidatorSetChanges fixes from Bucky's review update comment [ci skip] test TxEventBuffer update changelog fix TestValidatorSetChanges (2nd attempt) only do wg.Done when no errors benchmark event bus create pubsub server inside NewEventBus only expose config params (later if needed) set buffer capacity to 0 so we are not testing cache new tx event format: key = "Tx" plus a tag {"tx.hash": XYZ} This should allow to subscribe to all transactions! or a specific one using a query: "tm.events.type = Tx and tx.hash = '013ABF99434...'" use TimeoutCommit instead of afterPublishEventNewBlockTimeout TimeoutCommit is the time a node waits after committing a block, before it goes into the next height. So it will finish everything from the last block, but then wait a bit. The idea is this gives it time to hear more votes from other validators, to strengthen the commit it includes in the next block. But it also gives it time to hear about new transactions. waitForBlockWithUpdatedVals rewrite WAL crash tests Task: test that we can recover from any WAL crash. Solution: the old tests were relying on event hub being run in the same thread (we were injecting the private validator's last signature). when considering a rewrite, we considered two possible solutions: write a "fuzzy" testing system where WAL is crashing upon receiving a new message, or inject failures and trigger them in tests using something like https://github.com/coreos/gofail. remove sleep no cs.Lock around wal.Save test different cases (empty block, non-empty block, ...) comments add comments test 4 cases: empty block, non-empty block, non-empty block with smaller part size, many blocks fixes as per Bucky's last review reset subscriptions on UnsubscribeAll use a simple counter to track message for which we panicked also, set a smaller part size for all test cases
2017-06-26 08:00:30 -07:00
"context"
2015-12-01 20:12:01 -08:00
"fmt"
2016-11-23 15:20:46 -08:00
"io/ioutil"
2017-01-12 11:44:42 -08:00
"os"
"path"
2015-12-01 20:12:01 -08:00
"sort"
"sync"
2015-12-01 20:12:01 -08:00
"testing"
"time"
abcicli "github.com/tendermint/abci/client"
abci "github.com/tendermint/abci/types"
2015-12-01 20:12:01 -08:00
bc "github.com/tendermint/tendermint/blockchain"
2017-05-01 21:43:49 -07:00
cfg "github.com/tendermint/tendermint/config"
cstypes "github.com/tendermint/tendermint/consensus/types"
2015-12-01 20:12:01 -08:00
mempl "github.com/tendermint/tendermint/mempool"
2017-04-08 19:04:06 -07:00
"github.com/tendermint/tendermint/p2p"
2015-12-01 20:12:01 -08:00
sm "github.com/tendermint/tendermint/state"
"github.com/tendermint/tendermint/types"
2017-10-04 13:40:45 -07:00
cmn "github.com/tendermint/tmlibs/common"
2017-04-08 19:04:06 -07:00
dbm "github.com/tendermint/tmlibs/db"
2017-05-02 00:53:32 -07:00
"github.com/tendermint/tmlibs/log"
2015-12-01 20:12:01 -08:00
2017-01-12 12:53:32 -08:00
"github.com/tendermint/abci/example/counter"
"github.com/tendermint/abci/example/dummy"
"github.com/go-kit/kit/log/term"
2015-12-01 20:12:01 -08:00
)
new pubsub package comment out failing consensus tests for now rewrite rpc httpclient to use new pubsub package import pubsub as tmpubsub, query as tmquery make event IDs constants EventKey -> EventTypeKey rename EventsPubsub to PubSub mempool does not use pubsub rename eventsSub to pubsub new subscribe API fix channel size issues and consensus tests bugs refactor rpc client add missing discardFromChan method add mutex rename pubsub to eventBus remove IsRunning from WSRPCConnection interface (not needed) add a comment in broadcastNewRoundStepsAndVotes rename registerEventCallbacks to broadcastNewRoundStepsAndVotes See https://dave.cheney.net/2014/03/19/channel-axioms stop eventBuses after reactor tests remove unnecessary Unsubscribe return subscribe helper function move discardFromChan to where it is used subscribe now returns an err this gives us ability to refuse to subscribe if pubsub is at its max capacity. use context for control overflow cache queries handle err when subscribing in replay_test rename testClientID to testSubscriber extract var set channel buffer capacity to 1 in replay_file fix byzantine_test unsubscribe from single event, not all events refactor httpclient to return events to appropriate channels return failing testReplayCrashBeforeWriteVote test fix TestValidatorSetChanges refactor code a bit fix testReplayCrashBeforeWriteVote add comment fix TestValidatorSetChanges fixes from Bucky's review update comment [ci skip] test TxEventBuffer update changelog fix TestValidatorSetChanges (2nd attempt) only do wg.Done when no errors benchmark event bus create pubsub server inside NewEventBus only expose config params (later if needed) set buffer capacity to 0 so we are not testing cache new tx event format: key = "Tx" plus a tag {"tx.hash": XYZ} This should allow to subscribe to all transactions! or a specific one using a query: "tm.events.type = Tx and tx.hash = '013ABF99434...'" use TimeoutCommit instead of afterPublishEventNewBlockTimeout TimeoutCommit is the time a node waits after committing a block, before it goes into the next height. So it will finish everything from the last block, but then wait a bit. The idea is this gives it time to hear more votes from other validators, to strengthen the commit it includes in the next block. But it also gives it time to hear about new transactions. waitForBlockWithUpdatedVals rewrite WAL crash tests Task: test that we can recover from any WAL crash. Solution: the old tests were relying on event hub being run in the same thread (we were injecting the private validator's last signature). when considering a rewrite, we considered two possible solutions: write a "fuzzy" testing system where WAL is crashing upon receiving a new message, or inject failures and trigger them in tests using something like https://github.com/coreos/gofail. remove sleep no cs.Lock around wal.Save test different cases (empty block, non-empty block, ...) comments add comments test 4 cases: empty block, non-empty block, non-empty block with smaller part size, many blocks fixes as per Bucky's last review reset subscriptions on UnsubscribeAll use a simple counter to track message for which we panicked also, set a smaller part size for all test cases
2017-06-26 08:00:30 -07:00
const (
testSubscriber = "test-client"
)
// genesis, chain_id, priv_val
2017-05-04 19:33:08 -07:00
var config *cfg.Config // NOTE: must be reset for each _test.go file
2017-08-08 13:35:25 -07:00
var ensureTimeout = time.Second * 2
2017-01-12 11:44:42 -08:00
func ensureDir(dir string, mode os.FileMode) {
2017-10-04 13:40:45 -07:00
if err := cmn.EnsureDir(dir, mode); err != nil {
2017-01-12 11:44:42 -08:00
panic(err)
}
}
2017-05-04 19:33:08 -07:00
func ResetConfig(name string) *cfg.Config {
return cfg.ResetTestRoot(name)
2017-05-01 21:43:49 -07:00
}
//-------------------------------------------------------------------------------
// validator stub (a dummy consensus peer we control)
2015-12-01 20:12:01 -08:00
type validatorStub struct {
Index int // Validator index. NOTE: we don't assume validator set changes.
Height int64
2015-12-01 20:12:01 -08:00
Round int
2017-09-18 20:16:14 -07:00
types.PrivValidator
2015-12-01 20:12:01 -08:00
}
2017-12-01 22:47:55 -08:00
var testMinPower int64 = 10
2016-11-23 15:20:46 -08:00
2017-09-18 20:16:14 -07:00
func NewValidatorStub(privValidator types.PrivValidator, valIndex int) *validatorStub {
2015-12-01 20:12:01 -08:00
return &validatorStub{
Index: valIndex,
2015-12-01 20:12:01 -08:00
PrivValidator: privValidator,
}
}
func (vs *validatorStub) signVote(voteType byte, hash []byte, header types.PartSetHeader) (*types.Vote, error) {
vote := &types.Vote{
ValidatorIndex: vs.Index,
ValidatorAddress: vs.PrivValidator.GetAddress(),
2015-12-01 20:12:01 -08:00
Height: vs.Height,
Round: vs.Round,
Type: voteType,
2016-08-16 14:59:19 -07:00
BlockID: types.BlockID{hash, header},
2015-12-01 20:12:01 -08:00
}
2017-05-01 21:43:49 -07:00
err := vs.PrivValidator.SignVote(config.ChainID, vote)
2015-12-01 20:12:01 -08:00
return vote, err
}
// Sign vote for type/hash/header
2015-12-01 20:12:01 -08:00
func signVote(vs *validatorStub, voteType byte, hash []byte, header types.PartSetHeader) *types.Vote {
v, err := vs.signVote(voteType, hash, header)
if err != nil {
panic(fmt.Errorf("failed to sign vote: %v", err))
}
return v
}
func signVotes(voteType byte, hash []byte, header types.PartSetHeader, vss ...*validatorStub) []*types.Vote {
votes := make([]*types.Vote, len(vss))
for i, vs := range vss {
votes[i] = signVote(vs, voteType, hash, header)
}
return votes
}
func incrementHeight(vss ...*validatorStub) {
for _, vs := range vss {
vs.Height += 1
}
}
func incrementRound(vss ...*validatorStub) {
for _, vs := range vss {
vs.Round += 1
}
}
//-------------------------------------------------------------------------------
// Functions for transitioning the consensus state
func startTestRound(cs *ConsensusState, height int64, round int) {
cs.enterNewRound(height, round)
cs.startRoutines(0)
}
// Create proposal block from cs1 but sign it with vs
func decideProposal(cs1 *ConsensusState, vs *validatorStub, height int64, round int) (proposal *types.Proposal, block *types.Block) {
2015-12-01 20:12:01 -08:00
block, blockParts := cs1.createProposalBlock()
if block == nil { // on error
panic("error creating proposal block")
}
// Make proposal
polRound, polBlockID := cs1.Votes.POLInfo()
proposal = types.NewProposal(height, round, blockParts.Header(), polRound, polBlockID)
2017-05-01 21:43:49 -07:00
if err := vs.SignProposal(config.ChainID, proposal); err != nil {
2015-12-01 20:12:01 -08:00
panic(err)
}
return
}
func addVotes(to *ConsensusState, votes ...*types.Vote) {
for _, vote := range votes {
to.peerMsgQueue <- msgInfo{Msg: &VoteMessage{vote}}
2015-12-01 20:12:01 -08:00
}
}
func signAddVotes(to *ConsensusState, voteType byte, hash []byte, header types.PartSetHeader, vss ...*validatorStub) {
votes := signVotes(voteType, hash, header, vss...)
addVotes(to, votes...)
2015-12-01 20:12:01 -08:00
}
func validatePrevote(t *testing.T, cs *ConsensusState, round int, privVal *validatorStub, blockHash []byte) {
prevotes := cs.Votes.Prevotes(round)
var vote *types.Vote
if vote = prevotes.GetByAddress(privVal.GetAddress()); vote == nil {
2015-12-01 20:12:01 -08:00
panic("Failed to find prevote from validator")
}
if blockHash == nil {
2016-08-16 14:59:19 -07:00
if vote.BlockID.Hash != nil {
panic(fmt.Sprintf("Expected prevote to be for nil, got %X", vote.BlockID.Hash))
2015-12-01 20:12:01 -08:00
}
} else {
2016-08-16 14:59:19 -07:00
if !bytes.Equal(vote.BlockID.Hash, blockHash) {
panic(fmt.Sprintf("Expected prevote to be for %X, got %X", blockHash, vote.BlockID.Hash))
2015-12-01 20:12:01 -08:00
}
}
}
2015-12-13 11:56:05 -08:00
func validateLastPrecommit(t *testing.T, cs *ConsensusState, privVal *validatorStub, blockHash []byte) {
votes := cs.LastCommit
var vote *types.Vote
if vote = votes.GetByAddress(privVal.GetAddress()); vote == nil {
2015-12-13 11:56:05 -08:00
panic("Failed to find precommit from validator")
2015-12-01 20:12:01 -08:00
}
2016-08-16 14:59:19 -07:00
if !bytes.Equal(vote.BlockID.Hash, blockHash) {
panic(fmt.Sprintf("Expected precommit to be for %X, got %X", blockHash, vote.BlockID.Hash))
2015-12-01 20:12:01 -08:00
}
}
func validatePrecommit(t *testing.T, cs *ConsensusState, thisRound, lockRound int, privVal *validatorStub, votedBlockHash, lockedBlockHash []byte) {
precommits := cs.Votes.Precommits(thisRound)
var vote *types.Vote
if vote = precommits.GetByAddress(privVal.GetAddress()); vote == nil {
2015-12-01 20:12:01 -08:00
panic("Failed to find precommit from validator")
}
if votedBlockHash == nil {
2016-08-16 14:59:19 -07:00
if vote.BlockID.Hash != nil {
2015-12-01 20:12:01 -08:00
panic("Expected precommit to be for nil")
}
} else {
2016-08-16 14:59:19 -07:00
if !bytes.Equal(vote.BlockID.Hash, votedBlockHash) {
2015-12-01 20:12:01 -08:00
panic("Expected precommit to be for proposal block")
}
}
if lockedBlockHash == nil {
if cs.LockedRound != lockRound || cs.LockedBlock != nil {
panic(fmt.Sprintf("Expected to be locked on nil at round %d. Got locked at round %d with block %v", lockRound, cs.LockedRound, cs.LockedBlock))
}
} else {
if cs.LockedRound != lockRound || !bytes.Equal(cs.LockedBlock.Hash(), lockedBlockHash) {
panic(fmt.Sprintf("Expected block to be locked on round %d, got %d. Got locked block %X, expected %X", lockRound, cs.LockedRound, cs.LockedBlock.Hash(), lockedBlockHash))
}
}
}
func validatePrevoteAndPrecommit(t *testing.T, cs *ConsensusState, thisRound, lockRound int, privVal *validatorStub, votedBlockHash, lockedBlockHash []byte) {
// verify the prevote
validatePrevote(t, cs, thisRound, privVal, votedBlockHash)
// verify precommit
cs.mtx.Lock()
validatePrecommit(t, cs, thisRound, lockRound, privVal, votedBlockHash, lockedBlockHash)
cs.mtx.Unlock()
}
// genesis
func subscribeToVoter(cs *ConsensusState, addr []byte) chan interface{} {
new pubsub package comment out failing consensus tests for now rewrite rpc httpclient to use new pubsub package import pubsub as tmpubsub, query as tmquery make event IDs constants EventKey -> EventTypeKey rename EventsPubsub to PubSub mempool does not use pubsub rename eventsSub to pubsub new subscribe API fix channel size issues and consensus tests bugs refactor rpc client add missing discardFromChan method add mutex rename pubsub to eventBus remove IsRunning from WSRPCConnection interface (not needed) add a comment in broadcastNewRoundStepsAndVotes rename registerEventCallbacks to broadcastNewRoundStepsAndVotes See https://dave.cheney.net/2014/03/19/channel-axioms stop eventBuses after reactor tests remove unnecessary Unsubscribe return subscribe helper function move discardFromChan to where it is used subscribe now returns an err this gives us ability to refuse to subscribe if pubsub is at its max capacity. use context for control overflow cache queries handle err when subscribing in replay_test rename testClientID to testSubscriber extract var set channel buffer capacity to 1 in replay_file fix byzantine_test unsubscribe from single event, not all events refactor httpclient to return events to appropriate channels return failing testReplayCrashBeforeWriteVote test fix TestValidatorSetChanges refactor code a bit fix testReplayCrashBeforeWriteVote add comment fix TestValidatorSetChanges fixes from Bucky's review update comment [ci skip] test TxEventBuffer update changelog fix TestValidatorSetChanges (2nd attempt) only do wg.Done when no errors benchmark event bus create pubsub server inside NewEventBus only expose config params (later if needed) set buffer capacity to 0 so we are not testing cache new tx event format: key = "Tx" plus a tag {"tx.hash": XYZ} This should allow to subscribe to all transactions! or a specific one using a query: "tm.events.type = Tx and tx.hash = '013ABF99434...'" use TimeoutCommit instead of afterPublishEventNewBlockTimeout TimeoutCommit is the time a node waits after committing a block, before it goes into the next height. So it will finish everything from the last block, but then wait a bit. The idea is this gives it time to hear more votes from other validators, to strengthen the commit it includes in the next block. But it also gives it time to hear about new transactions. waitForBlockWithUpdatedVals rewrite WAL crash tests Task: test that we can recover from any WAL crash. Solution: the old tests were relying on event hub being run in the same thread (we were injecting the private validator's last signature). when considering a rewrite, we considered two possible solutions: write a "fuzzy" testing system where WAL is crashing upon receiving a new message, or inject failures and trigger them in tests using something like https://github.com/coreos/gofail. remove sleep no cs.Lock around wal.Save test different cases (empty block, non-empty block, ...) comments add comments test 4 cases: empty block, non-empty block, non-empty block with smaller part size, many blocks fixes as per Bucky's last review reset subscriptions on UnsubscribeAll use a simple counter to track message for which we panicked also, set a smaller part size for all test cases
2017-06-26 08:00:30 -07:00
voteCh0 := make(chan interface{})
err := cs.eventBus.Subscribe(context.Background(), testSubscriber, types.EventQueryVote, voteCh0)
if err != nil {
panic(fmt.Sprintf("failed to subscribe %s to %v", testSubscriber, types.EventQueryVote))
}
voteCh := make(chan interface{})
go func() {
new pubsub package comment out failing consensus tests for now rewrite rpc httpclient to use new pubsub package import pubsub as tmpubsub, query as tmquery make event IDs constants EventKey -> EventTypeKey rename EventsPubsub to PubSub mempool does not use pubsub rename eventsSub to pubsub new subscribe API fix channel size issues and consensus tests bugs refactor rpc client add missing discardFromChan method add mutex rename pubsub to eventBus remove IsRunning from WSRPCConnection interface (not needed) add a comment in broadcastNewRoundStepsAndVotes rename registerEventCallbacks to broadcastNewRoundStepsAndVotes See https://dave.cheney.net/2014/03/19/channel-axioms stop eventBuses after reactor tests remove unnecessary Unsubscribe return subscribe helper function move discardFromChan to where it is used subscribe now returns an err this gives us ability to refuse to subscribe if pubsub is at its max capacity. use context for control overflow cache queries handle err when subscribing in replay_test rename testClientID to testSubscriber extract var set channel buffer capacity to 1 in replay_file fix byzantine_test unsubscribe from single event, not all events refactor httpclient to return events to appropriate channels return failing testReplayCrashBeforeWriteVote test fix TestValidatorSetChanges refactor code a bit fix testReplayCrashBeforeWriteVote add comment fix TestValidatorSetChanges fixes from Bucky's review update comment [ci skip] test TxEventBuffer update changelog fix TestValidatorSetChanges (2nd attempt) only do wg.Done when no errors benchmark event bus create pubsub server inside NewEventBus only expose config params (later if needed) set buffer capacity to 0 so we are not testing cache new tx event format: key = "Tx" plus a tag {"tx.hash": XYZ} This should allow to subscribe to all transactions! or a specific one using a query: "tm.events.type = Tx and tx.hash = '013ABF99434...'" use TimeoutCommit instead of afterPublishEventNewBlockTimeout TimeoutCommit is the time a node waits after committing a block, before it goes into the next height. So it will finish everything from the last block, but then wait a bit. The idea is this gives it time to hear more votes from other validators, to strengthen the commit it includes in the next block. But it also gives it time to hear about new transactions. waitForBlockWithUpdatedVals rewrite WAL crash tests Task: test that we can recover from any WAL crash. Solution: the old tests were relying on event hub being run in the same thread (we were injecting the private validator's last signature). when considering a rewrite, we considered two possible solutions: write a "fuzzy" testing system where WAL is crashing upon receiving a new message, or inject failures and trigger them in tests using something like https://github.com/coreos/gofail. remove sleep no cs.Lock around wal.Save test different cases (empty block, non-empty block, ...) comments add comments test 4 cases: empty block, non-empty block, non-empty block with smaller part size, many blocks fixes as per Bucky's last review reset subscriptions on UnsubscribeAll use a simple counter to track message for which we panicked also, set a smaller part size for all test cases
2017-06-26 08:00:30 -07:00
for v := range voteCh0 {
2017-04-28 14:57:06 -07:00
vote := v.(types.TMEventData).Unwrap().(types.EventDataVote)
// we only fire for our own votes
if bytes.Equal(addr, vote.Vote.ValidatorAddress) {
voteCh <- v
}
}
}()
return voteCh
2016-01-18 12:57:57 -08:00
}
2015-12-01 20:12:01 -08:00
//-------------------------------------------------------------------------------
// consensus states
2017-09-18 20:16:14 -07:00
func newConsensusState(state *sm.State, pv types.PrivValidator, app abci.Application) *ConsensusState {
return newConsensusStateWithConfig(config, state, pv, app)
2016-10-11 09:51:48 -07:00
}
2017-09-18 20:16:14 -07:00
func newConsensusStateWithConfig(thisConfig *cfg.Config, state *sm.State, pv types.PrivValidator, app abci.Application) *ConsensusState {
2015-12-01 20:12:01 -08:00
blockDB := dbm.NewMemDB()
new pubsub package comment out failing consensus tests for now rewrite rpc httpclient to use new pubsub package import pubsub as tmpubsub, query as tmquery make event IDs constants EventKey -> EventTypeKey rename EventsPubsub to PubSub mempool does not use pubsub rename eventsSub to pubsub new subscribe API fix channel size issues and consensus tests bugs refactor rpc client add missing discardFromChan method add mutex rename pubsub to eventBus remove IsRunning from WSRPCConnection interface (not needed) add a comment in broadcastNewRoundStepsAndVotes rename registerEventCallbacks to broadcastNewRoundStepsAndVotes See https://dave.cheney.net/2014/03/19/channel-axioms stop eventBuses after reactor tests remove unnecessary Unsubscribe return subscribe helper function move discardFromChan to where it is used subscribe now returns an err this gives us ability to refuse to subscribe if pubsub is at its max capacity. use context for control overflow cache queries handle err when subscribing in replay_test rename testClientID to testSubscriber extract var set channel buffer capacity to 1 in replay_file fix byzantine_test unsubscribe from single event, not all events refactor httpclient to return events to appropriate channels return failing testReplayCrashBeforeWriteVote test fix TestValidatorSetChanges refactor code a bit fix testReplayCrashBeforeWriteVote add comment fix TestValidatorSetChanges fixes from Bucky's review update comment [ci skip] test TxEventBuffer update changelog fix TestValidatorSetChanges (2nd attempt) only do wg.Done when no errors benchmark event bus create pubsub server inside NewEventBus only expose config params (later if needed) set buffer capacity to 0 so we are not testing cache new tx event format: key = "Tx" plus a tag {"tx.hash": XYZ} This should allow to subscribe to all transactions! or a specific one using a query: "tm.events.type = Tx and tx.hash = '013ABF99434...'" use TimeoutCommit instead of afterPublishEventNewBlockTimeout TimeoutCommit is the time a node waits after committing a block, before it goes into the next height. So it will finish everything from the last block, but then wait a bit. The idea is this gives it time to hear more votes from other validators, to strengthen the commit it includes in the next block. But it also gives it time to hear about new transactions. waitForBlockWithUpdatedVals rewrite WAL crash tests Task: test that we can recover from any WAL crash. Solution: the old tests were relying on event hub being run in the same thread (we were injecting the private validator's last signature). when considering a rewrite, we considered two possible solutions: write a "fuzzy" testing system where WAL is crashing upon receiving a new message, or inject failures and trigger them in tests using something like https://github.com/coreos/gofail. remove sleep no cs.Lock around wal.Save test different cases (empty block, non-empty block, ...) comments add comments test 4 cases: empty block, non-empty block, non-empty block with smaller part size, many blocks fixes as per Bucky's last review reset subscriptions on UnsubscribeAll use a simple counter to track message for which we panicked also, set a smaller part size for all test cases
2017-06-26 08:00:30 -07:00
return newConsensusStateWithConfigAndBlockStore(thisConfig, state, pv, app, blockDB)
}
func newConsensusStateWithConfigAndBlockStore(thisConfig *cfg.Config, state *sm.State, pv types.PrivValidator, app abci.Application, blockDB dbm.DB) *ConsensusState {
// Get BlockStore
2015-12-01 20:12:01 -08:00
blockStore := bc.NewBlockStore(blockDB)
// one for mempool, one for consensus
mtx := new(sync.Mutex)
2017-01-12 12:53:32 -08:00
proxyAppConnMem := abcicli.NewLocalClient(mtx, app)
proxyAppConnCon := abcicli.NewLocalClient(mtx, app)
2015-12-01 20:12:01 -08:00
// Make Mempool
mempool := mempl.NewMempool(thisConfig.Mempool, proxyAppConnMem, 0)
2017-05-02 00:53:32 -07:00
mempool.SetLogger(log.TestingLogger().With("module", "mempool"))
if thisConfig.Consensus.WaitForTxs() {
mempool.EnableTxsAvailable()
}
2015-12-01 20:12:01 -08:00
// Make ConsensusReactor
2017-05-01 21:43:49 -07:00
cs := NewConsensusState(thisConfig.Consensus, state, proxyAppConnCon, blockStore, mempool)
2017-05-02 00:53:32 -07:00
cs.SetLogger(log.TestingLogger())
2016-01-18 12:57:57 -08:00
cs.SetPrivValidator(pv)
2015-12-01 20:12:01 -08:00
new pubsub package comment out failing consensus tests for now rewrite rpc httpclient to use new pubsub package import pubsub as tmpubsub, query as tmquery make event IDs constants EventKey -> EventTypeKey rename EventsPubsub to PubSub mempool does not use pubsub rename eventsSub to pubsub new subscribe API fix channel size issues and consensus tests bugs refactor rpc client add missing discardFromChan method add mutex rename pubsub to eventBus remove IsRunning from WSRPCConnection interface (not needed) add a comment in broadcastNewRoundStepsAndVotes rename registerEventCallbacks to broadcastNewRoundStepsAndVotes See https://dave.cheney.net/2014/03/19/channel-axioms stop eventBuses after reactor tests remove unnecessary Unsubscribe return subscribe helper function move discardFromChan to where it is used subscribe now returns an err this gives us ability to refuse to subscribe if pubsub is at its max capacity. use context for control overflow cache queries handle err when subscribing in replay_test rename testClientID to testSubscriber extract var set channel buffer capacity to 1 in replay_file fix byzantine_test unsubscribe from single event, not all events refactor httpclient to return events to appropriate channels return failing testReplayCrashBeforeWriteVote test fix TestValidatorSetChanges refactor code a bit fix testReplayCrashBeforeWriteVote add comment fix TestValidatorSetChanges fixes from Bucky's review update comment [ci skip] test TxEventBuffer update changelog fix TestValidatorSetChanges (2nd attempt) only do wg.Done when no errors benchmark event bus create pubsub server inside NewEventBus only expose config params (later if needed) set buffer capacity to 0 so we are not testing cache new tx event format: key = "Tx" plus a tag {"tx.hash": XYZ} This should allow to subscribe to all transactions! or a specific one using a query: "tm.events.type = Tx and tx.hash = '013ABF99434...'" use TimeoutCommit instead of afterPublishEventNewBlockTimeout TimeoutCommit is the time a node waits after committing a block, before it goes into the next height. So it will finish everything from the last block, but then wait a bit. The idea is this gives it time to hear more votes from other validators, to strengthen the commit it includes in the next block. But it also gives it time to hear about new transactions. waitForBlockWithUpdatedVals rewrite WAL crash tests Task: test that we can recover from any WAL crash. Solution: the old tests were relying on event hub being run in the same thread (we were injecting the private validator's last signature). when considering a rewrite, we considered two possible solutions: write a "fuzzy" testing system where WAL is crashing upon receiving a new message, or inject failures and trigger them in tests using something like https://github.com/coreos/gofail. remove sleep no cs.Lock around wal.Save test different cases (empty block, non-empty block, ...) comments add comments test 4 cases: empty block, non-empty block, non-empty block with smaller part size, many blocks fixes as per Bucky's last review reset subscriptions on UnsubscribeAll use a simple counter to track message for which we panicked also, set a smaller part size for all test cases
2017-06-26 08:00:30 -07:00
eventBus := types.NewEventBus()
eventBus.SetLogger(log.TestingLogger().With("module", "events"))
eventBus.Start()
cs.SetEventBus(eventBus)
2016-01-18 12:57:57 -08:00
return cs
}
2015-12-01 20:12:01 -08:00
2017-09-18 20:16:14 -07:00
func loadPrivValidator(config *cfg.Config) *types.PrivValidatorFS {
2017-05-04 19:33:08 -07:00
privValidatorFile := config.PrivValidatorFile()
2017-01-12 11:44:42 -08:00
ensureDir(path.Dir(privValidatorFile), 0700)
2017-09-18 20:16:14 -07:00
privValidator := types.LoadOrGenPrivValidatorFS(privValidatorFile)
privValidator.Reset()
return privValidator
}
new pubsub package comment out failing consensus tests for now rewrite rpc httpclient to use new pubsub package import pubsub as tmpubsub, query as tmquery make event IDs constants EventKey -> EventTypeKey rename EventsPubsub to PubSub mempool does not use pubsub rename eventsSub to pubsub new subscribe API fix channel size issues and consensus tests bugs refactor rpc client add missing discardFromChan method add mutex rename pubsub to eventBus remove IsRunning from WSRPCConnection interface (not needed) add a comment in broadcastNewRoundStepsAndVotes rename registerEventCallbacks to broadcastNewRoundStepsAndVotes See https://dave.cheney.net/2014/03/19/channel-axioms stop eventBuses after reactor tests remove unnecessary Unsubscribe return subscribe helper function move discardFromChan to where it is used subscribe now returns an err this gives us ability to refuse to subscribe if pubsub is at its max capacity. use context for control overflow cache queries handle err when subscribing in replay_test rename testClientID to testSubscriber extract var set channel buffer capacity to 1 in replay_file fix byzantine_test unsubscribe from single event, not all events refactor httpclient to return events to appropriate channels return failing testReplayCrashBeforeWriteVote test fix TestValidatorSetChanges refactor code a bit fix testReplayCrashBeforeWriteVote add comment fix TestValidatorSetChanges fixes from Bucky's review update comment [ci skip] test TxEventBuffer update changelog fix TestValidatorSetChanges (2nd attempt) only do wg.Done when no errors benchmark event bus create pubsub server inside NewEventBus only expose config params (later if needed) set buffer capacity to 0 so we are not testing cache new tx event format: key = "Tx" plus a tag {"tx.hash": XYZ} This should allow to subscribe to all transactions! or a specific one using a query: "tm.events.type = Tx and tx.hash = '013ABF99434...'" use TimeoutCommit instead of afterPublishEventNewBlockTimeout TimeoutCommit is the time a node waits after committing a block, before it goes into the next height. So it will finish everything from the last block, but then wait a bit. The idea is this gives it time to hear more votes from other validators, to strengthen the commit it includes in the next block. But it also gives it time to hear about new transactions. waitForBlockWithUpdatedVals rewrite WAL crash tests Task: test that we can recover from any WAL crash. Solution: the old tests were relying on event hub being run in the same thread (we were injecting the private validator's last signature). when considering a rewrite, we considered two possible solutions: write a "fuzzy" testing system where WAL is crashing upon receiving a new message, or inject failures and trigger them in tests using something like https://github.com/coreos/gofail. remove sleep no cs.Lock around wal.Save test different cases (empty block, non-empty block, ...) comments add comments test 4 cases: empty block, non-empty block, non-empty block with smaller part size, many blocks fixes as per Bucky's last review reset subscriptions on UnsubscribeAll use a simple counter to track message for which we panicked also, set a smaller part size for all test cases
2017-06-26 08:00:30 -07:00
func fixedConsensusStateDummy(config *cfg.Config, logger log.Logger) *ConsensusState {
stateDB := dbm.NewMemDB()
state, _ := sm.MakeGenesisStateFromFile(stateDB, config.GenesisFile())
new pubsub package comment out failing consensus tests for now rewrite rpc httpclient to use new pubsub package import pubsub as tmpubsub, query as tmquery make event IDs constants EventKey -> EventTypeKey rename EventsPubsub to PubSub mempool does not use pubsub rename eventsSub to pubsub new subscribe API fix channel size issues and consensus tests bugs refactor rpc client add missing discardFromChan method add mutex rename pubsub to eventBus remove IsRunning from WSRPCConnection interface (not needed) add a comment in broadcastNewRoundStepsAndVotes rename registerEventCallbacks to broadcastNewRoundStepsAndVotes See https://dave.cheney.net/2014/03/19/channel-axioms stop eventBuses after reactor tests remove unnecessary Unsubscribe return subscribe helper function move discardFromChan to where it is used subscribe now returns an err this gives us ability to refuse to subscribe if pubsub is at its max capacity. use context for control overflow cache queries handle err when subscribing in replay_test rename testClientID to testSubscriber extract var set channel buffer capacity to 1 in replay_file fix byzantine_test unsubscribe from single event, not all events refactor httpclient to return events to appropriate channels return failing testReplayCrashBeforeWriteVote test fix TestValidatorSetChanges refactor code a bit fix testReplayCrashBeforeWriteVote add comment fix TestValidatorSetChanges fixes from Bucky's review update comment [ci skip] test TxEventBuffer update changelog fix TestValidatorSetChanges (2nd attempt) only do wg.Done when no errors benchmark event bus create pubsub server inside NewEventBus only expose config params (later if needed) set buffer capacity to 0 so we are not testing cache new tx event format: key = "Tx" plus a tag {"tx.hash": XYZ} This should allow to subscribe to all transactions! or a specific one using a query: "tm.events.type = Tx and tx.hash = '013ABF99434...'" use TimeoutCommit instead of afterPublishEventNewBlockTimeout TimeoutCommit is the time a node waits after committing a block, before it goes into the next height. So it will finish everything from the last block, but then wait a bit. The idea is this gives it time to hear more votes from other validators, to strengthen the commit it includes in the next block. But it also gives it time to hear about new transactions. waitForBlockWithUpdatedVals rewrite WAL crash tests Task: test that we can recover from any WAL crash. Solution: the old tests were relying on event hub being run in the same thread (we were injecting the private validator's last signature). when considering a rewrite, we considered two possible solutions: write a "fuzzy" testing system where WAL is crashing upon receiving a new message, or inject failures and trigger them in tests using something like https://github.com/coreos/gofail. remove sleep no cs.Lock around wal.Save test different cases (empty block, non-empty block, ...) comments add comments test 4 cases: empty block, non-empty block, non-empty block with smaller part size, many blocks fixes as per Bucky's last review reset subscriptions on UnsubscribeAll use a simple counter to track message for which we panicked also, set a smaller part size for all test cases
2017-06-26 08:00:30 -07:00
state.SetLogger(logger.With("module", "state"))
privValidator := loadPrivValidator(config)
cs := newConsensusState(state, privValidator, dummy.NewDummyApplication())
new pubsub package comment out failing consensus tests for now rewrite rpc httpclient to use new pubsub package import pubsub as tmpubsub, query as tmquery make event IDs constants EventKey -> EventTypeKey rename EventsPubsub to PubSub mempool does not use pubsub rename eventsSub to pubsub new subscribe API fix channel size issues and consensus tests bugs refactor rpc client add missing discardFromChan method add mutex rename pubsub to eventBus remove IsRunning from WSRPCConnection interface (not needed) add a comment in broadcastNewRoundStepsAndVotes rename registerEventCallbacks to broadcastNewRoundStepsAndVotes See https://dave.cheney.net/2014/03/19/channel-axioms stop eventBuses after reactor tests remove unnecessary Unsubscribe return subscribe helper function move discardFromChan to where it is used subscribe now returns an err this gives us ability to refuse to subscribe if pubsub is at its max capacity. use context for control overflow cache queries handle err when subscribing in replay_test rename testClientID to testSubscriber extract var set channel buffer capacity to 1 in replay_file fix byzantine_test unsubscribe from single event, not all events refactor httpclient to return events to appropriate channels return failing testReplayCrashBeforeWriteVote test fix TestValidatorSetChanges refactor code a bit fix testReplayCrashBeforeWriteVote add comment fix TestValidatorSetChanges fixes from Bucky's review update comment [ci skip] test TxEventBuffer update changelog fix TestValidatorSetChanges (2nd attempt) only do wg.Done when no errors benchmark event bus create pubsub server inside NewEventBus only expose config params (later if needed) set buffer capacity to 0 so we are not testing cache new tx event format: key = "Tx" plus a tag {"tx.hash": XYZ} This should allow to subscribe to all transactions! or a specific one using a query: "tm.events.type = Tx and tx.hash = '013ABF99434...'" use TimeoutCommit instead of afterPublishEventNewBlockTimeout TimeoutCommit is the time a node waits after committing a block, before it goes into the next height. So it will finish everything from the last block, but then wait a bit. The idea is this gives it time to hear more votes from other validators, to strengthen the commit it includes in the next block. But it also gives it time to hear about new transactions. waitForBlockWithUpdatedVals rewrite WAL crash tests Task: test that we can recover from any WAL crash. Solution: the old tests were relying on event hub being run in the same thread (we were injecting the private validator's last signature). when considering a rewrite, we considered two possible solutions: write a "fuzzy" testing system where WAL is crashing upon receiving a new message, or inject failures and trigger them in tests using something like https://github.com/coreos/gofail. remove sleep no cs.Lock around wal.Save test different cases (empty block, non-empty block, ...) comments add comments test 4 cases: empty block, non-empty block, non-empty block with smaller part size, many blocks fixes as per Bucky's last review reset subscriptions on UnsubscribeAll use a simple counter to track message for which we panicked also, set a smaller part size for all test cases
2017-06-26 08:00:30 -07:00
cs.SetLogger(logger)
return cs
2016-09-13 19:25:11 -07:00
}
2016-01-18 12:57:57 -08:00
func randConsensusState(nValidators int) (*ConsensusState, []*validatorStub) {
// Get State
state, privVals := randGenesisState(nValidators, false, 10)
vss := make([]*validatorStub, nValidators)
2015-12-01 20:12:01 -08:00
cs := newConsensusState(state, privVals[0], counter.NewCounterApplication(true))
2017-05-02 00:53:32 -07:00
cs.SetLogger(log.TestingLogger())
2015-12-08 13:00:59 -08:00
2015-12-01 20:12:01 -08:00
for i := 0; i < nValidators; i++ {
vss[i] = NewValidatorStub(privVals[i], i)
2015-12-01 20:12:01 -08:00
}
// since cs1 starts at 1
incrementHeight(vss[1:]...)
return cs, vss
}
//-------------------------------------------------------------------------------
new pubsub package comment out failing consensus tests for now rewrite rpc httpclient to use new pubsub package import pubsub as tmpubsub, query as tmquery make event IDs constants EventKey -> EventTypeKey rename EventsPubsub to PubSub mempool does not use pubsub rename eventsSub to pubsub new subscribe API fix channel size issues and consensus tests bugs refactor rpc client add missing discardFromChan method add mutex rename pubsub to eventBus remove IsRunning from WSRPCConnection interface (not needed) add a comment in broadcastNewRoundStepsAndVotes rename registerEventCallbacks to broadcastNewRoundStepsAndVotes See https://dave.cheney.net/2014/03/19/channel-axioms stop eventBuses after reactor tests remove unnecessary Unsubscribe return subscribe helper function move discardFromChan to where it is used subscribe now returns an err this gives us ability to refuse to subscribe if pubsub is at its max capacity. use context for control overflow cache queries handle err when subscribing in replay_test rename testClientID to testSubscriber extract var set channel buffer capacity to 1 in replay_file fix byzantine_test unsubscribe from single event, not all events refactor httpclient to return events to appropriate channels return failing testReplayCrashBeforeWriteVote test fix TestValidatorSetChanges refactor code a bit fix testReplayCrashBeforeWriteVote add comment fix TestValidatorSetChanges fixes from Bucky's review update comment [ci skip] test TxEventBuffer update changelog fix TestValidatorSetChanges (2nd attempt) only do wg.Done when no errors benchmark event bus create pubsub server inside NewEventBus only expose config params (later if needed) set buffer capacity to 0 so we are not testing cache new tx event format: key = "Tx" plus a tag {"tx.hash": XYZ} This should allow to subscribe to all transactions! or a specific one using a query: "tm.events.type = Tx and tx.hash = '013ABF99434...'" use TimeoutCommit instead of afterPublishEventNewBlockTimeout TimeoutCommit is the time a node waits after committing a block, before it goes into the next height. So it will finish everything from the last block, but then wait a bit. The idea is this gives it time to hear more votes from other validators, to strengthen the commit it includes in the next block. But it also gives it time to hear about new transactions. waitForBlockWithUpdatedVals rewrite WAL crash tests Task: test that we can recover from any WAL crash. Solution: the old tests were relying on event hub being run in the same thread (we were injecting the private validator's last signature). when considering a rewrite, we considered two possible solutions: write a "fuzzy" testing system where WAL is crashing upon receiving a new message, or inject failures and trigger them in tests using something like https://github.com/coreos/gofail. remove sleep no cs.Lock around wal.Save test different cases (empty block, non-empty block, ...) comments add comments test 4 cases: empty block, non-empty block, non-empty block with smaller part size, many blocks fixes as per Bucky's last review reset subscriptions on UnsubscribeAll use a simple counter to track message for which we panicked also, set a smaller part size for all test cases
2017-06-26 08:00:30 -07:00
func ensureNoNewStep(stepCh <-chan interface{}) {
2017-08-08 13:35:25 -07:00
timer := time.NewTimer(ensureTimeout)
select {
2017-07-13 12:03:19 -07:00
case <-timer.C:
break
case <-stepCh:
2017-07-13 12:03:19 -07:00
panic("We should be stuck waiting, not moving to the next step")
}
}
new pubsub package comment out failing consensus tests for now rewrite rpc httpclient to use new pubsub package import pubsub as tmpubsub, query as tmquery make event IDs constants EventKey -> EventTypeKey rename EventsPubsub to PubSub mempool does not use pubsub rename eventsSub to pubsub new subscribe API fix channel size issues and consensus tests bugs refactor rpc client add missing discardFromChan method add mutex rename pubsub to eventBus remove IsRunning from WSRPCConnection interface (not needed) add a comment in broadcastNewRoundStepsAndVotes rename registerEventCallbacks to broadcastNewRoundStepsAndVotes See https://dave.cheney.net/2014/03/19/channel-axioms stop eventBuses after reactor tests remove unnecessary Unsubscribe return subscribe helper function move discardFromChan to where it is used subscribe now returns an err this gives us ability to refuse to subscribe if pubsub is at its max capacity. use context for control overflow cache queries handle err when subscribing in replay_test rename testClientID to testSubscriber extract var set channel buffer capacity to 1 in replay_file fix byzantine_test unsubscribe from single event, not all events refactor httpclient to return events to appropriate channels return failing testReplayCrashBeforeWriteVote test fix TestValidatorSetChanges refactor code a bit fix testReplayCrashBeforeWriteVote add comment fix TestValidatorSetChanges fixes from Bucky's review update comment [ci skip] test TxEventBuffer update changelog fix TestValidatorSetChanges (2nd attempt) only do wg.Done when no errors benchmark event bus create pubsub server inside NewEventBus only expose config params (later if needed) set buffer capacity to 0 so we are not testing cache new tx event format: key = "Tx" plus a tag {"tx.hash": XYZ} This should allow to subscribe to all transactions! or a specific one using a query: "tm.events.type = Tx and tx.hash = '013ABF99434...'" use TimeoutCommit instead of afterPublishEventNewBlockTimeout TimeoutCommit is the time a node waits after committing a block, before it goes into the next height. So it will finish everything from the last block, but then wait a bit. The idea is this gives it time to hear more votes from other validators, to strengthen the commit it includes in the next block. But it also gives it time to hear about new transactions. waitForBlockWithUpdatedVals rewrite WAL crash tests Task: test that we can recover from any WAL crash. Solution: the old tests were relying on event hub being run in the same thread (we were injecting the private validator's last signature). when considering a rewrite, we considered two possible solutions: write a "fuzzy" testing system where WAL is crashing upon receiving a new message, or inject failures and trigger them in tests using something like https://github.com/coreos/gofail. remove sleep no cs.Lock around wal.Save test different cases (empty block, non-empty block, ...) comments add comments test 4 cases: empty block, non-empty block, non-empty block with smaller part size, many blocks fixes as per Bucky's last review reset subscriptions on UnsubscribeAll use a simple counter to track message for which we panicked also, set a smaller part size for all test cases
2017-06-26 08:00:30 -07:00
func ensureNewStep(stepCh <-chan interface{}) {
2017-08-08 13:35:25 -07:00
timer := time.NewTimer(ensureTimeout)
2017-07-13 12:03:19 -07:00
select {
case <-timer.C:
panic("We shouldnt be stuck waiting")
case <-stepCh:
break
}
}
//-------------------------------------------------------------------------------
// consensus nets
// consensusLogger is a TestingLogger which uses a different
// color for each validator ("validator" key must exist).
func consensusLogger() log.Logger {
return log.TestingLoggerWithColorFn(func(keyvals ...interface{}) term.FgBgColor {
for i := 0; i < len(keyvals)-1; i += 2 {
if keyvals[i] == "validator" {
return term.FgBgColor{Fg: term.Color(uint8(keyvals[i+1].(int) + 1))}
}
}
return term.FgBgColor{}
})
}
2017-08-09 22:09:04 -07:00
func randConsensusNet(nValidators int, testName string, tickerFunc func() TimeoutTicker, appFunc func() abci.Application, configOpts ...func(*cfg.Config)) []*ConsensusState {
2016-06-25 21:40:53 -07:00
genDoc, privVals := randGenesisDoc(nValidators, false, 10)
css := make([]*ConsensusState, nValidators)
logger := consensusLogger()
2016-06-25 21:40:53 -07:00
for i := 0; i < nValidators; i++ {
db := dbm.NewMemDB() // each state needs its own db
state, _ := sm.MakeGenesisState(db, genDoc)
state.SetLogger(logger.With("module", "state", "validator", i))
2016-06-25 21:40:53 -07:00
state.Save()
2017-10-04 13:40:45 -07:00
thisConfig := ResetConfig(cmn.Fmt("%s_%d", testName, i))
2017-08-09 22:09:04 -07:00
for _, opt := range configOpts {
opt(thisConfig)
}
2017-05-04 19:33:08 -07:00
ensureDir(path.Dir(thisConfig.Consensus.WalFile()), 0700) // dir for wal
2017-10-23 07:03:54 -07:00
app := appFunc()
vals := types.TM2PB.Validators(state.Validators)
app.InitChain(abci.RequestInitChain{Validators: vals})
css[i] = newConsensusStateWithConfig(thisConfig, state, privVals[i], app)
css[i].SetLogger(logger.With("validator", i))
css[i].SetTimeoutTicker(tickerFunc())
2016-06-25 21:40:53 -07:00
}
return css
}
2016-11-23 15:20:46 -08:00
// nPeers = nValidators + nNotValidator
2017-01-12 12:53:32 -08:00
func randConsensusNetWithPeers(nValidators, nPeers int, testName string, tickerFunc func() TimeoutTicker, appFunc func() abci.Application) []*ConsensusState {
2017-12-01 22:47:55 -08:00
genDoc, privVals := randGenesisDoc(nValidators, false, testMinPower)
2016-11-23 15:20:46 -08:00
css := make([]*ConsensusState, nPeers)
new pubsub package comment out failing consensus tests for now rewrite rpc httpclient to use new pubsub package import pubsub as tmpubsub, query as tmquery make event IDs constants EventKey -> EventTypeKey rename EventsPubsub to PubSub mempool does not use pubsub rename eventsSub to pubsub new subscribe API fix channel size issues and consensus tests bugs refactor rpc client add missing discardFromChan method add mutex rename pubsub to eventBus remove IsRunning from WSRPCConnection interface (not needed) add a comment in broadcastNewRoundStepsAndVotes rename registerEventCallbacks to broadcastNewRoundStepsAndVotes See https://dave.cheney.net/2014/03/19/channel-axioms stop eventBuses after reactor tests remove unnecessary Unsubscribe return subscribe helper function move discardFromChan to where it is used subscribe now returns an err this gives us ability to refuse to subscribe if pubsub is at its max capacity. use context for control overflow cache queries handle err when subscribing in replay_test rename testClientID to testSubscriber extract var set channel buffer capacity to 1 in replay_file fix byzantine_test unsubscribe from single event, not all events refactor httpclient to return events to appropriate channels return failing testReplayCrashBeforeWriteVote test fix TestValidatorSetChanges refactor code a bit fix testReplayCrashBeforeWriteVote add comment fix TestValidatorSetChanges fixes from Bucky's review update comment [ci skip] test TxEventBuffer update changelog fix TestValidatorSetChanges (2nd attempt) only do wg.Done when no errors benchmark event bus create pubsub server inside NewEventBus only expose config params (later if needed) set buffer capacity to 0 so we are not testing cache new tx event format: key = "Tx" plus a tag {"tx.hash": XYZ} This should allow to subscribe to all transactions! or a specific one using a query: "tm.events.type = Tx and tx.hash = '013ABF99434...'" use TimeoutCommit instead of afterPublishEventNewBlockTimeout TimeoutCommit is the time a node waits after committing a block, before it goes into the next height. So it will finish everything from the last block, but then wait a bit. The idea is this gives it time to hear more votes from other validators, to strengthen the commit it includes in the next block. But it also gives it time to hear about new transactions. waitForBlockWithUpdatedVals rewrite WAL crash tests Task: test that we can recover from any WAL crash. Solution: the old tests were relying on event hub being run in the same thread (we were injecting the private validator's last signature). when considering a rewrite, we considered two possible solutions: write a "fuzzy" testing system where WAL is crashing upon receiving a new message, or inject failures and trigger them in tests using something like https://github.com/coreos/gofail. remove sleep no cs.Lock around wal.Save test different cases (empty block, non-empty block, ...) comments add comments test 4 cases: empty block, non-empty block, non-empty block with smaller part size, many blocks fixes as per Bucky's last review reset subscriptions on UnsubscribeAll use a simple counter to track message for which we panicked also, set a smaller part size for all test cases
2017-06-26 08:00:30 -07:00
logger := consensusLogger()
2016-11-23 15:20:46 -08:00
for i := 0; i < nPeers; i++ {
db := dbm.NewMemDB() // each state needs its own db
state, _ := sm.MakeGenesisState(db, genDoc)
new pubsub package comment out failing consensus tests for now rewrite rpc httpclient to use new pubsub package import pubsub as tmpubsub, query as tmquery make event IDs constants EventKey -> EventTypeKey rename EventsPubsub to PubSub mempool does not use pubsub rename eventsSub to pubsub new subscribe API fix channel size issues and consensus tests bugs refactor rpc client add missing discardFromChan method add mutex rename pubsub to eventBus remove IsRunning from WSRPCConnection interface (not needed) add a comment in broadcastNewRoundStepsAndVotes rename registerEventCallbacks to broadcastNewRoundStepsAndVotes See https://dave.cheney.net/2014/03/19/channel-axioms stop eventBuses after reactor tests remove unnecessary Unsubscribe return subscribe helper function move discardFromChan to where it is used subscribe now returns an err this gives us ability to refuse to subscribe if pubsub is at its max capacity. use context for control overflow cache queries handle err when subscribing in replay_test rename testClientID to testSubscriber extract var set channel buffer capacity to 1 in replay_file fix byzantine_test unsubscribe from single event, not all events refactor httpclient to return events to appropriate channels return failing testReplayCrashBeforeWriteVote test fix TestValidatorSetChanges refactor code a bit fix testReplayCrashBeforeWriteVote add comment fix TestValidatorSetChanges fixes from Bucky's review update comment [ci skip] test TxEventBuffer update changelog fix TestValidatorSetChanges (2nd attempt) only do wg.Done when no errors benchmark event bus create pubsub server inside NewEventBus only expose config params (later if needed) set buffer capacity to 0 so we are not testing cache new tx event format: key = "Tx" plus a tag {"tx.hash": XYZ} This should allow to subscribe to all transactions! or a specific one using a query: "tm.events.type = Tx and tx.hash = '013ABF99434...'" use TimeoutCommit instead of afterPublishEventNewBlockTimeout TimeoutCommit is the time a node waits after committing a block, before it goes into the next height. So it will finish everything from the last block, but then wait a bit. The idea is this gives it time to hear more votes from other validators, to strengthen the commit it includes in the next block. But it also gives it time to hear about new transactions. waitForBlockWithUpdatedVals rewrite WAL crash tests Task: test that we can recover from any WAL crash. Solution: the old tests were relying on event hub being run in the same thread (we were injecting the private validator's last signature). when considering a rewrite, we considered two possible solutions: write a "fuzzy" testing system where WAL is crashing upon receiving a new message, or inject failures and trigger them in tests using something like https://github.com/coreos/gofail. remove sleep no cs.Lock around wal.Save test different cases (empty block, non-empty block, ...) comments add comments test 4 cases: empty block, non-empty block, non-empty block with smaller part size, many blocks fixes as per Bucky's last review reset subscriptions on UnsubscribeAll use a simple counter to track message for which we panicked also, set a smaller part size for all test cases
2017-06-26 08:00:30 -07:00
state.SetLogger(logger.With("module", "state", "validator", i))
2016-11-23 15:20:46 -08:00
state.Save()
2017-10-04 13:40:45 -07:00
thisConfig := ResetConfig(cmn.Fmt("%s_%d", testName, i))
2017-05-04 19:33:08 -07:00
ensureDir(path.Dir(thisConfig.Consensus.WalFile()), 0700) // dir for wal
2017-09-18 20:16:14 -07:00
var privVal types.PrivValidator
2016-11-23 15:20:46 -08:00
if i < nValidators {
privVal = privVals[i]
} else {
2017-10-04 13:40:45 -07:00
_, tempFilePath := cmn.Tempfile("priv_validator_")
2017-09-18 20:16:14 -07:00
privVal = types.GenPrivValidatorFS(tempFilePath)
2016-11-23 15:20:46 -08:00
}
2017-10-23 07:03:54 -07:00
app := appFunc()
vals := types.TM2PB.Validators(state.Validators)
app.InitChain(abci.RequestInitChain{Validators: vals})
css[i] = newConsensusStateWithConfig(thisConfig, state, privVal, app)
new pubsub package comment out failing consensus tests for now rewrite rpc httpclient to use new pubsub package import pubsub as tmpubsub, query as tmquery make event IDs constants EventKey -> EventTypeKey rename EventsPubsub to PubSub mempool does not use pubsub rename eventsSub to pubsub new subscribe API fix channel size issues and consensus tests bugs refactor rpc client add missing discardFromChan method add mutex rename pubsub to eventBus remove IsRunning from WSRPCConnection interface (not needed) add a comment in broadcastNewRoundStepsAndVotes rename registerEventCallbacks to broadcastNewRoundStepsAndVotes See https://dave.cheney.net/2014/03/19/channel-axioms stop eventBuses after reactor tests remove unnecessary Unsubscribe return subscribe helper function move discardFromChan to where it is used subscribe now returns an err this gives us ability to refuse to subscribe if pubsub is at its max capacity. use context for control overflow cache queries handle err when subscribing in replay_test rename testClientID to testSubscriber extract var set channel buffer capacity to 1 in replay_file fix byzantine_test unsubscribe from single event, not all events refactor httpclient to return events to appropriate channels return failing testReplayCrashBeforeWriteVote test fix TestValidatorSetChanges refactor code a bit fix testReplayCrashBeforeWriteVote add comment fix TestValidatorSetChanges fixes from Bucky's review update comment [ci skip] test TxEventBuffer update changelog fix TestValidatorSetChanges (2nd attempt) only do wg.Done when no errors benchmark event bus create pubsub server inside NewEventBus only expose config params (later if needed) set buffer capacity to 0 so we are not testing cache new tx event format: key = "Tx" plus a tag {"tx.hash": XYZ} This should allow to subscribe to all transactions! or a specific one using a query: "tm.events.type = Tx and tx.hash = '013ABF99434...'" use TimeoutCommit instead of afterPublishEventNewBlockTimeout TimeoutCommit is the time a node waits after committing a block, before it goes into the next height. So it will finish everything from the last block, but then wait a bit. The idea is this gives it time to hear more votes from other validators, to strengthen the commit it includes in the next block. But it also gives it time to hear about new transactions. waitForBlockWithUpdatedVals rewrite WAL crash tests Task: test that we can recover from any WAL crash. Solution: the old tests were relying on event hub being run in the same thread (we were injecting the private validator's last signature). when considering a rewrite, we considered two possible solutions: write a "fuzzy" testing system where WAL is crashing upon receiving a new message, or inject failures and trigger them in tests using something like https://github.com/coreos/gofail. remove sleep no cs.Lock around wal.Save test different cases (empty block, non-empty block, ...) comments add comments test 4 cases: empty block, non-empty block, non-empty block with smaller part size, many blocks fixes as per Bucky's last review reset subscriptions on UnsubscribeAll use a simple counter to track message for which we panicked also, set a smaller part size for all test cases
2017-06-26 08:00:30 -07:00
css[i].SetLogger(logger.With("validator", i))
css[i].SetTimeoutTicker(tickerFunc())
2016-11-23 15:20:46 -08:00
}
return css
}
2017-09-12 17:49:22 -07:00
func getSwitchIndex(switches []*p2p.Switch, peer p2p.Peer) int {
for i, s := range switches {
2017-09-12 17:49:22 -07:00
if bytes.Equal(peer.NodeInfo().PubKey.Address(), s.NodeInfo().PubKey.Address()) {
return i
2016-07-11 17:40:48 -07:00
}
}
panic("didnt find peer in switches")
return -1
2016-07-11 17:40:48 -07:00
}
//-------------------------------------------------------------------------------
// genesis
2015-12-01 20:12:01 -08:00
2017-09-18 20:16:14 -07:00
func randGenesisDoc(numValidators int, randPower bool, minPower int64) (*types.GenesisDoc, []*types.PrivValidatorFS) {
2015-12-01 20:12:01 -08:00
validators := make([]types.GenesisValidator, numValidators)
2017-09-18 20:16:14 -07:00
privValidators := make([]*types.PrivValidatorFS, numValidators)
2015-12-01 20:12:01 -08:00
for i := 0; i < numValidators; i++ {
val, privVal := types.RandValidator(randPower, minPower)
validators[i] = types.GenesisValidator{
PubKey: val.PubKey,
2017-09-21 11:37:34 -07:00
Power: val.VotingPower,
2015-12-01 20:12:01 -08:00
}
privValidators[i] = privVal
}
sort.Sort(types.PrivValidatorsByAddress(privValidators))
return &types.GenesisDoc{
GenesisTime: time.Now(),
2017-05-01 21:43:49 -07:00
ChainID: config.ChainID,
2015-12-01 20:12:01 -08:00
Validators: validators,
}, privValidators
}
2016-11-23 15:20:46 -08:00
2017-09-18 20:16:14 -07:00
func randGenesisState(numValidators int, randPower bool, minPower int64) (*sm.State, []*types.PrivValidatorFS) {
genDoc, privValidators := randGenesisDoc(numValidators, randPower, minPower)
db := dbm.NewMemDB()
s0, _ := sm.MakeGenesisState(db, genDoc)
2017-05-02 00:53:32 -07:00
s0.SetLogger(log.TestingLogger().With("module", "state"))
s0.Save()
return s0, privValidators
2016-11-23 15:20:46 -08:00
}
2016-12-06 16:54:10 -08:00
//------------------------------------
// mock ticker
func newMockTickerFunc(onlyOnce bool) func() TimeoutTicker {
return func() TimeoutTicker {
return &mockTicker{
c: make(chan timeoutInfo, 10),
onlyOnce: onlyOnce,
}
}
}
2016-12-22 18:51:58 -08:00
// mock ticker only fires on RoundStepNewHeight
// and only once if onlyOnce=true
type mockTicker struct {
c chan timeoutInfo
mtx sync.Mutex
onlyOnce bool
fired bool
}
func (m *mockTicker) Start() error {
return nil
}
func (m *mockTicker) Stop() error {
return nil
}
func (m *mockTicker) ScheduleTimeout(ti timeoutInfo) {
m.mtx.Lock()
defer m.mtx.Unlock()
if m.onlyOnce && m.fired {
return
}
if ti.Step == cstypes.RoundStepNewHeight {
m.c <- ti
m.fired = true
}
}
func (m *mockTicker) Chan() <-chan timeoutInfo {
return m.c
}
2017-05-12 14:07:53 -07:00
func (mockTicker) SetLogger(log.Logger) {
}
//------------------------------------
2017-01-12 12:53:32 -08:00
func newCounter() abci.Application {
return counter.NewCounterApplication(true)
}
2017-01-12 12:53:32 -08:00
func newPersistentDummy() abci.Application {
dir, _ := ioutil.TempDir("/tmp", "persistent-dummy")
return dummy.NewPersistentDummyApplication(dir)
}