// NewStorePool creates a StorePool and registers the store updating callback
// with gossip.
func NewStorePool(
ambient log.AmbientContext,
g *gossip.Gossip,
clock *hlc.Clock,
rpcContext *rpc.Context,
timeUntilStoreDead time.Duration,
stopper *stop.Stopper,
deterministic bool,
) *StorePool {
sp := &StorePool{
AmbientContext: ambient,
clock: clock,
timeUntilStoreDead: timeUntilStoreDead,
rpcContext: rpcContext,
failedReservationsTimeout: envutil.EnvOrDefaultDuration("COCKROACH_FAILED_RESERVATION_TIMEOUT",
defaultFailedReservationsTimeout),
declinedReservationsTimeout: envutil.EnvOrDefaultDuration("COCKROACH_DECLINED_RESERVATION_TIMEOUT",
defaultDeclinedReservationsTimeout),
resolver: GossipAddressResolver(g),
deterministic: deterministic,
}
sp.mu.storeDetails = make(map[roachpb.StoreID]*storeDetail)
heap.Init(&sp.mu.queue)
sp.mu.nodeLocalities = make(map[roachpb.NodeID]roachpb.Locality)
storeRegex := gossip.MakePrefixPattern(gossip.KeyStorePrefix)
g.RegisterCallback(storeRegex, sp.storeGossipUpdate)
deadReplicasRegex := gossip.MakePrefixPattern(gossip.KeyDeadReplicasPrefix)
g.RegisterCallback(deadReplicasRegex, sp.deadReplicasGossipUpdate)
sp.start(stopper)
return sp
}
func initFakeTime() {
if offset := envutil.EnvOrDefaultDuration("COCKROACH_SIMULATED_OFFSET", 0); offset == 0 {
nowFunc = time.Now
} else {
nowFunc = func() time.Time {
return time.Now().Add(offset)
}
}
}
func (r *RocksDB) open() error {
var ver storageVersion
if len(r.dir) != 0 {
log.Infof(context.TODO(), "opening rocksdb instance at %q", r.dir)
// Check the version number.
var err error
if ver, err = getVersion(r.dir); err != nil {
return err
}
if ver < versionMinimum || ver > versionCurrent {
// Instead of an error, we should call a migration if possible when
// one is needed immediately following the DBOpen call.
return fmt.Errorf("incompatible rocksdb data version, current:%d, on disk:%d, minimum:%d",
versionCurrent, ver, versionMinimum)
}
} else {
if log.V(2) {
log.Infof(context.TODO(), "opening in memory rocksdb instance")
}
// In memory dbs are always current.
ver = versionCurrent
}
blockSize := envutil.EnvOrDefaultBytes("COCKROACH_ROCKSDB_BLOCK_SIZE", defaultBlockSize)
walTTL := envutil.EnvOrDefaultDuration("COCKROACH_ROCKSDB_WAL_TTL", 0).Seconds()
status := C.DBOpen(&r.rdb, goToCSlice([]byte(r.dir)),
C.DBOptions{
cache: r.cache.cache,
block_size: C.uint64_t(blockSize),
wal_ttl_seconds: C.uint64_t(walTTL),
allow_os_buffer: C.bool(true),
logging_enabled: C.bool(log.V(3)),
num_cpu: C.int(runtime.NumCPU()),
max_open_files: C.int(r.maxOpenFiles),
})
if err := statusToError(status); err != nil {
return errors.Errorf("could not open rocksdb instance: %s", err)
}
// Update or add the version file if needed.
if ver < versionCurrent {
if err := writeVersionFile(r.dir); err != nil {
return err
}
}
// Start a goroutine that will finish when the underlying handle
// is deallocated. This is used to check a leak in tests.
go func() {
<-r.deallocated
}()
return nil
}
// readEnvironmentVariables populates all context values that are environment
// variable based. Note that this only happens when initializing a node and not
// when NewContext is called.
func (cfg *Config) readEnvironmentVariables() {
// cockroach-linearizable
cfg.Linearizable = envutil.EnvOrDefaultBool("COCKROACH_LINEARIZABLE", cfg.Linearizable)
cfg.ConsistencyCheckPanicOnFailure = envutil.EnvOrDefaultBool("COCKROACH_CONSISTENCY_CHECK_PANIC_ON_FAILURE", cfg.ConsistencyCheckPanicOnFailure)
cfg.MaxOffset = envutil.EnvOrDefaultDuration("COCKROACH_MAX_OFFSET", cfg.MaxOffset)
cfg.MetricsSampleInterval = envutil.EnvOrDefaultDuration("COCKROACH_METRICS_SAMPLE_INTERVAL", cfg.MetricsSampleInterval)
cfg.ScanInterval = envutil.EnvOrDefaultDuration("COCKROACH_SCAN_INTERVAL", cfg.ScanInterval)
cfg.ScanMaxIdleTime = envutil.EnvOrDefaultDuration("COCKROACH_SCAN_MAX_IDLE_TIME", cfg.ScanMaxIdleTime)
cfg.TimeUntilStoreDead = envutil.EnvOrDefaultDuration("COCKROACH_TIME_UNTIL_STORE_DEAD", cfg.TimeUntilStoreDead)
cfg.ConsistencyCheckInterval = envutil.EnvOrDefaultDuration("COCKROACH_CONSISTENCY_CHECK_INTERVAL", cfg.ConsistencyCheckInterval)
}
// startGossip loops on a periodic ticker to gossip node-related
// information. Starts a goroutine to loop until the node is closed.
func (n *Node) startGossip(stopper *stop.Stopper) {
stopper.RunWorker(func() {
ctx := n.AnnotateCtx(context.Background())
// This should always return immediately and acts as a sanity check that we
// don't try to gossip before we're connected.
select {
case <-n.storeCfg.Gossip.Connected:
default:
panic(fmt.Sprintf("%s: not connected to gossip", n))
}
// Verify we've already gossiped our node descriptor.
if _, err := n.storeCfg.Gossip.GetNodeDescriptor(n.Descriptor.NodeID); err != nil {
panic(err)
}
gossipStoresInterval := envutil.EnvOrDefaultDuration("COCKROACH_GOSSIP_STORES_INTERVAL",
gossip.DefaultGossipStoresInterval)
statusTicker := time.NewTicker(gossipStatusInterval)
storesTicker := time.NewTicker(gossipStoresInterval)
nodeTicker := time.NewTicker(gossipNodeDescriptorInterval)
defer storesTicker.Stop()
defer nodeTicker.Stop()
n.gossipStores(ctx) // one-off run before going to sleep
for {
select {
case <-statusTicker.C:
n.storeCfg.Gossip.LogStatus()
case <-storesTicker.C:
n.gossipStores(ctx)
case <-nodeTicker.C:
if err := n.storeCfg.Gossip.SetNodeDescriptor(&n.Descriptor); err != nil {
log.Warningf(ctx, "couldn't gossip descriptor for node %d: %s", n.Descriptor.NodeID, err)
}
case <-stopper.ShouldStop():
return
}
}
})
}
)
// Gossip metrics counter names.
var (
MetaConnectionsIncomingGauge = metric.Metadata{Name: "gossip.connections.incoming"}
MetaConnectionsOutgoingGauge = metric.Metadata{Name: "gossip.connections.outgoing"}
MetaConnectionsRefusedRates = metric.Metadata{Name: "gossip.connections.refused"}
MetaInfosSentRates = metric.Metadata{Name: "gossip.infos.sent"}
MetaInfosReceivedRates = metric.Metadata{Name: "gossip.infos.received"}
MetaBytesSentRates = metric.Metadata{Name: "gossip.bytes.sent"}
MetaBytesReceivedRates = metric.Metadata{Name: "gossip.bytes.received"}
)
var (
// GossipStoresInterval is the interval for gossipping storage-related info.
GossipStoresInterval = envutil.EnvOrDefaultDuration("COCKROACH_GOSSIP_STORES_INTERVAL",
defaultGossipStoresInterval)
)
// Storage is an interface which allows the gossip instance
// to read and write bootstrapping data to persistent storage
// between instantiations.
type Storage interface {
// ReadBootstrapInfo fetches the bootstrap data from the persistent
// store into the provided bootstrap protobuf. Returns nil or an
// error on failure.
ReadBootstrapInfo(*BootstrapInfo) error
// WriteBootstrapInfo stores the provided bootstrap data to the
// persistent store. Returns nil or an error on failure.
WriteBootstrapInfo(*BootstrapInfo) error
}
"github.com/cockroachdb/cockroach/pkg/util/log"
"github.com/cockroachdb/cockroach/pkg/util/retry"
"github.com/cockroachdb/cockroach/pkg/util/timeutil"
"github.com/cockroachdb/cockroach/pkg/util/tracing"
basictracer "github.com/opentracing/basictracer-go"
opentracing "github.com/opentracing/opentracing-go"
"github.com/pkg/errors"
)
// COCKROACH_TRACE_SQL=duration can be used to log SQL transactions that take
// longer than duration to complete. For example, COCKROACH_TRACE_SQL=1s will
// log the trace for any transaction that takes 1s or longer. To log traces for
// all transactions use COCKROACH_TRACE_SQL=1ns. Note that any positive
// duration will enable tracing and will slow down all execution because traces
// are gathered for all transactions even if they are not output.
var traceSQLDuration = envutil.EnvOrDefaultDuration("COCKROACH_TRACE_SQL", 0)
var traceSQL = traceSQLDuration > 0
// COCKROACH_TRACE_7881 can be used to trace all SQL transactions, in the hope
// that we'll catch #7881 and dump the current trace for debugging.
var traceSQLFor7881 = envutil.EnvOrDefaultBool("COCKROACH_TRACE_7881", false)
// span baggage key used for marking a span
const keyFor7881Sample = "found#7881"
// Session contains the state of a SQL client connection.
// Create instances using NewSession().
type Session struct {
Database string
// SearchPath is a list of databases that will be searched for a table name
// before the database. Currently, this is used only for SELECTs.