featurebase/disco/disco.go
Seebs b3a4e52a13 simplify, streamline, and possibly debug embedded etcd
The root problem this is attempting to address is sporadic
weird cases in which etcd mistakenly thinks it's down even when
it's up. I am not confident that this is addressed, but there's
a reasonable chance that it is, and I can't trigger it at the
moment, but it was always sporadic, so that doesn't prove much.

There's a lot going on here, and it comes into roughly three
categories.

First: Dropping unused/unneeded code. There's a lot of leftover
bits from the initial development and refactoring of this.

Second: Unifying and shuffling some of the design. We had
multiple interfaces which are functionally impossible to
usefully implement separately, so they're combined together,
and in some cases, moved.

Third: Streamlining logic and simplifying design choices.

This is combined into one commit because the changes are
thoroughly entertwined with each other and you can't usefully
break most of them out.

Also, a bunch of test coverage for most of these changes.

Big changes:

We merge the topology and disco packages.  The topology and disco
packages being separate creates a complicated tangle of problems
and dependencies.  The fundamental problem, approximately, is that
topology.Node has to track disco.NodeState.

There's three core interfaces interacting here:
	topology.Noder (maintains list of nodes)
	disco.Stator (maintains the state of a node)
	disco.Metadator (stores, possibly retrieves, node metadata)
But the node state mantained by the Noder *is* the set of node
metadata, plus state updates produced by Stators. The only actual
non-trivial and usable implementation of these interfaces is a single
thing which implements all three, and in which the implementations
share a single backend data source which they are all modifying.

But you can't move Noder into disco, because Noder has to refer
to topology.Node, but topology.Node refers to disco.

Solution: First, merge these two packages. Second, merge these
three interfaces, to provide a single interface which is more
clear about the fact that (metadator.)SetMetadata() and
(stator.)Started() are both changing the output we'll get from
(noder.)Nodes().

We rework the node state tracking.

We have this nodeStates map which is almost unused. Really, we
don't need it at all. Every node's state is either its last heartbeat
state or "Unknown", so we simplify this a bit. Also, we ensure that
the populateNodeStates function itself is yielding the sorted nodes
list, so we don't have to be as worried about possible later lookups
of sortedNodes happening outside a lock. We also add diagnostics
for deleting nodes from the metadata list (this should never happen),
and try to track heartbeat state more closely.

This is *probably* what fixes the underlying reported problem,
if anything did.

Still an open issue: Make heartbeat state changes aware of when
they're talking about *this* node and possibly not try to
mark it down? Except this may have a flaw: That would result in
each node disagreeing with other nodes in etcd about the state
of that node in the failure cases, and undermine the point of
using etcd to keep these states consistent.

We reduce the number of contexts and cancelfuncs in the etcd wrapper.

We create a shared context for the non-etcd.embed children of our
etcd wrapper, the heartbeat/keepalive and the node watcher, so we
can cancel that one context and cancel all of those at once, so
we don't need to separately track a function to call to cancel
the watch, AND be closing another channel. Also, our shutdown
now propagates automatically to the various etcd API calls we've
made for things like the node watcher and keepalive calls.

We still need to watch that channel in watchNodesOnce, though,
because apparently the watch doesn't yield an error even if the
context calling it is canceled. Whee.

This should reduce the risk of ending up in an inconsistent state,
and also the Close() function is probably idempotent now.

Smaller changes:

* Remove config-generators that existed to generate etcd
  configs but were used only for tests that no longer exist
  or make sense.
* Move the logic to generate etcd configs into the etcd
  package, instead of the "testing" subpackage. This allows
  us to write a self-contained config generator for
  clusters where the nodes know about each other, but do
  this just with etcd, not with full featurebase servers.
* Move the thing generating `fake:%d` socket names into
  the etcd package, which is the only place we use it.
  Also simplify it slightly.
* Don't panic on invalid URLs, report errors from them.
* At least try to use etcd's config.Validate functionality.
  It's underdocumented, so we're not sure what it will report,
  but at least if it does we'll get reports from it and
  know what they are?
* Try to handle CompactRevision errors from watches more
  correctly -- after a CompactRevision, any future attempt
  to watch from a lower revision will necessarily fail, so
  we adjust our target revision up. We don't have good
  testing for this.
* Drop the Metadata() method (that used to be in Metadator)
  because nothing ever used it and it didn't make much sense
  to try.
* Convert SetMetadata from taking an arbitrary json blob
  to taking the only data that would ever be valid since
  we always use it to extract node information anyway.
* Drop several unused functions, unexport things only used
  internally.
* Replace Started() with SetState("STARTED"), allowing us
  to write tests that mess with states. We weren't really thinking
  carefully about state transitions sometimes and now it's much
  easier to do that thinking.
* Stop leaving stray localhost:2380 and localhost:2379 in
  our embed config. We still sometimes see peer requests from
  those and I honestly don't know why, but at least it should
  be rarer.
2022-07-21 11:42:35 -05:00

428 lines
13 KiB
Go

// Copyright 2021 Molecula Corp. All rights reserved.
package disco
import (
"context"
"fmt"
"io"
"path"
"sync"
)
var (
ErrTooManyResults error = fmt.Errorf("too many results")
ErrNoResults error = fmt.Errorf("no results")
ErrKeyDeleted error = fmt.Errorf("key deleted")
ErrIndexExists error = fmt.Errorf("index already exists")
ErrIndexDoesNotExist error = fmt.Errorf("index does not exist")
ErrFieldExists error = fmt.Errorf("field already exists")
ErrFieldDoesNotExist error = fmt.Errorf("field does not exist")
ErrViewExists error = fmt.Errorf("view already exists")
ErrViewDoesNotExist error = fmt.Errorf("view does not exist")
ErrKeyDoesNotExist error = fmt.Errorf("key does not exist")
)
type Peer struct {
URL string
ID string
}
func (p *Peer) String() string {
return fmt.Sprintf(`{"ID": "%s", "URL": "%s"}`, p.ID, p.URL)
}
type DisCo interface {
io.Closer
Start(ctx context.Context) (InitialClusterState, error)
IsLeader() bool
ID() string
Leader() *Peer
Peers() []*Peer
DeleteNode(ctx context.Context, id string) error
}
type (
InitialClusterState string
// ClusterState represents the state returned in the /status endpoint.
ClusterState string
)
const (
InitialClusterStateNew InitialClusterState = "new"
InitialClusterStateExisting InitialClusterState = "existing"
ClusterStateUnknown ClusterState = "UNKNOWN" // default cluster state. It is returned when we are not able to get the real actual state.
ClusterStateStarting ClusterState = "STARTING" // cluster is starting and some internal services are not ready yet.
ClusterStateDegraded ClusterState = "DEGRADED" // cluster is running but we've lost some # of hosts >0 but < replicaN. Only read queries are allowed.
ClusterStateNormal ClusterState = "NORMAL" // cluster is up and running.
ClusterStateDown ClusterState = "DOWN" // cluster is unable to serve queries.
)
type NodeState string
const (
NodeStateUnknown NodeState = "UNKNOWN"
NodeStateStarting NodeState = "STARTING"
NodeStateStarted NodeState = "STARTED"
)
// Schema is a map of all indexes, each of those being a map of fields, then
// views.
type Schema map[string]*Index
// Index is a struct which contains the data encoded for the index as well as
// for each of its fields.
type Index struct {
Data []byte
Fields map[string]*Field
}
// Field is a struct which contains the data encoded for the field as well as
// for each of its views.
type Field struct {
Data []byte
Views map[string]struct{}
}
// Schemator is the source of truth for different schema elements.
// All nodes will store and retrieve information from the same source,
// having the same information at the same time.
type Schemator interface {
// Schema return the actual pilosa schema. If the schema is not present, an error is returned.
Schema(ctx context.Context) (Schema, error)
// Index gets a specific index data by name.
Index(ctx context.Context, name string) ([]byte, error)
CreateIndex(ctx context.Context, name string, val []byte) error
DeleteIndex(ctx context.Context, name string) error
Field(ctx context.Context, index, field string) ([]byte, error)
CreateField(ctx context.Context, index, field string, val []byte) error
UpdateField(ctx context.Context, index, field string, val []byte) error
DeleteField(ctx context.Context, index, field string) error
View(ctx context.Context, index, field, view string) (bool, error)
CreateView(ctx context.Context, index, field, view string) error
DeleteView(ctx context.Context, index, field, view string) error
}
// Sharder is an interface used to maintain the set of availableShards bitmaps
// per field.
type Sharder interface {
Shards(ctx context.Context, index, field string) ([][]byte, error)
SetShards(ctx context.Context, index, field string, shards []byte) error
}
// NopDisCo represents a DisCo that doesn't do anything.
var NopDisCo DisCo = &nopDisCo{}
type nopDisCo struct{}
// Close no-op.
func (n *nopDisCo) Close() error {
return nil
}
// Start is a no-op implementation of the DisCo Start method.
func (n *nopDisCo) Start(ctx context.Context) (InitialClusterState, error) {
return InitialClusterStateNew, nil
}
// ID is a no-op implementation of the DisCo ID method.
func (n *nopDisCo) ID() string {
return ""
}
// IsLeader is a no-op implementation of the DisCo IsLeader method.
func (n *nopDisCo) IsLeader() bool {
return false
}
// Leader is a no-op implementation of the DisCo Leader method.
func (n *nopDisCo) Leader() *Peer {
return nil
}
// Peers is a no-op implementation of the DisCo Peers method.
func (n *nopDisCo) Peers() []*Peer {
return nil
}
// DeleteNode a no-op implementation of the DisCo DeleteNode method.
func (n *nopDisCo) DeleteNode(context.Context, string) error {
return nil
}
// NopSharder represents a Sharder that doesn't do anything.
var NopSharder Sharder = &nopSharder{}
type nopSharder struct{}
// Shards is a no-op implementation of the Sharder Shards method.
func (n *nopSharder) Shards(ctx context.Context, index, field string) ([][]byte, error) {
return nil, nil
}
// AddShards is a no-op implementation of the Sharder AddShards method.
func (n *nopSharder) SetShards(ctx context.Context, index, field string, shards []byte) error {
return nil
}
// NopSchemator represents a Schemator that doesn't do anything.
var NopSchemator Schemator = &nopSchemator{}
type nopSchemator struct{}
// Schema is a no-op implementation of the Schemator Schema method.
func (*nopSchemator) Schema(ctx context.Context) (Schema, error) { return nil, nil }
// Index is a no-op implementation of the Schemator Index method.
func (*nopSchemator) Index(ctx context.Context, name string) ([]byte, error) { return nil, nil }
// CreateIndex is a no-op implementation of the Schemator CreateIndex method.
func (*nopSchemator) CreateIndex(ctx context.Context, name string, val []byte) error { return nil }
// DeleteIndex is a no-op implementation of the Schemator DeleteIndex method.
func (*nopSchemator) DeleteIndex(ctx context.Context, name string) error { return nil }
// Field is a no-op implementation of the Schemator Field method.
func (*nopSchemator) Field(ctx context.Context, index, field string) ([]byte, error) { return nil, nil }
// CreateField is a no-op implementation of the Schemator CreateField method.
func (*nopSchemator) CreateField(ctx context.Context, index, field string, val []byte) error {
return nil
}
// UpdateField is a no-op implementation of the Schemator UpdateField method.
func (*nopSchemator) UpdateField(ctx context.Context, index, field string, val []byte) error {
return nil
}
// DeleteField is a no-op implementation of the Schemator DeleteField method.
func (*nopSchemator) DeleteField(ctx context.Context, index, field string) error { return nil }
// View is a no-op implementation of the Schemator View method.
func (*nopSchemator) View(ctx context.Context, index, field, view string) (bool, error) {
return false, nil
}
// CreateView is a no-op implementation of the Schemator CreateView method.
func (*nopSchemator) CreateView(ctx context.Context, index, field, view string) error {
return nil
}
// DeleteView is a no-op implementation of the Schemator DeleteView method.
func (*nopSchemator) DeleteView(ctx context.Context, index, field, view string) error { return nil }
// InMemSchemator represents a Schemator that manages the schema in memory. The
// intention is that this would be used for testing.
var InMemSchemator Schemator = &inMemSchemator{
schema: make(Schema),
}
type inMemSchemator struct {
mu sync.RWMutex
schema Schema
}
// NewInMemSchemator instantiates an InMemSchemator
// this allows new holders to have thier own, and not rely on a shared instance
func NewInMemSchemator() *inMemSchemator {
return &inMemSchemator{
schema: make(Schema),
}
}
// Schema is an in-memory implementation of the Schemator Schema method.
func (s *inMemSchemator) Schema(ctx context.Context) (Schema, error) {
s.mu.RLock()
defer s.mu.RUnlock()
return s.schema, nil
}
// Index is an in-memory implementation of the Schemator Index method.
func (s *inMemSchemator) Index(ctx context.Context, name string) ([]byte, error) {
s.mu.RLock()
defer s.mu.RUnlock()
idx, ok := s.schema[name]
if !ok {
return nil, ErrIndexDoesNotExist
}
return idx.Data, nil
}
// CreateIndex is an in-memory implementation of the Schemator CreateIndex method.
func (s *inMemSchemator) CreateIndex(ctx context.Context, name string, val []byte) error {
s.mu.Lock()
defer s.mu.Unlock()
if idx, ok := s.schema[name]; ok {
// The current logic in pilosa doesn't allow us to return ErrIndexExists
// here, so for now we just update the Data value if the index already
// exists.
idx.Data = val
return nil
}
s.schema[name] = &Index{
Data: val,
Fields: make(map[string]*Field),
}
return nil
}
// DeleteIndex is an in-memory implementation of the Schemator DeleteIndex method.
func (s *inMemSchemator) DeleteIndex(ctx context.Context, name string) error {
s.mu.Lock()
defer s.mu.Unlock()
delete(s.schema, name)
return nil
}
// Field is an in-memory implementation of the Schemator Field method.
func (s *inMemSchemator) Field(ctx context.Context, index, field string) ([]byte, error) {
s.mu.RLock()
defer s.mu.RUnlock()
idx, ok := s.schema[index]
if !ok {
return nil, ErrIndexDoesNotExist
}
fld, ok := idx.Fields[field]
if !ok {
return nil, ErrFieldDoesNotExist
}
return fld.Data, nil
}
// CreateField is an in-memory implementation of the Schemator CreateField method.
func (s *inMemSchemator) CreateField(ctx context.Context, index, field string, val []byte) error {
s.mu.Lock()
defer s.mu.Unlock()
idx, ok := s.schema[index]
if !ok {
return ErrIndexDoesNotExist
}
if fld, ok := idx.Fields[field]; ok {
// The current logic in pilosa doesn't allow us to return ErrFieldExists
// here, so for now we just update the Data value if the field already
// exists.
fld.Data = val
return nil
}
idx.Fields[field] = &Field{
Data: val,
Views: make(map[string]struct{}),
}
return nil
}
func (s *inMemSchemator) UpdateField(ctx context.Context, index, field string, val []byte) error {
s.mu.Lock()
defer s.mu.Unlock()
idx, ok := s.schema[index]
if !ok {
return ErrIndexDoesNotExist
}
if fld, ok := idx.Fields[field]; ok {
// The current logic in pilosa doesn't allow us to return ErrFieldExists
// here, so for now we just update the Data value if the field already
// exists.
fld.Data = val
return nil
} else {
return ErrFieldDoesNotExist
}
}
// DeleteField is an in-memory implementation of the Schemator DeleteField method.
func (s *inMemSchemator) DeleteField(ctx context.Context, index, field string) error {
s.mu.Lock()
defer s.mu.Unlock()
idx, ok := s.schema[index]
if !ok {
return ErrIndexDoesNotExist
}
delete(idx.Fields, field)
return nil
}
// View is an in-memory implementation of the Schemator View method.
func (s *inMemSchemator) View(ctx context.Context, index, field, view string) (bool, error) {
s.mu.RLock()
defer s.mu.RUnlock()
idx, ok := s.schema[index]
if !ok {
return false, ErrIndexDoesNotExist
}
fld, ok := idx.Fields[field]
if !ok {
return false, ErrFieldDoesNotExist
}
_, ok = fld.Views[view]
return ok, nil
}
// CreateView is an in-memory implementation of the Schemator CreateView method.
func (s *inMemSchemator) CreateView(ctx context.Context, index, field, view string) error {
s.mu.Lock()
defer s.mu.Unlock()
idx, ok := s.schema[index]
if !ok {
return ErrIndexDoesNotExist
}
fld, ok := idx.Fields[field]
if !ok {
return ErrFieldDoesNotExist
}
// The current logic in pilosa doesn't allow us to return ErrViewExists
// here, so for now we just update the value if the view already exists.
fld.Views[view] = struct{}{}
return nil
}
// DeleteView is an in-memory implementation of the Schemator DeleteView method.
func (s *inMemSchemator) DeleteView(ctx context.Context, index, field, view string) error {
s.mu.Lock()
defer s.mu.Unlock()
idx, ok := s.schema[index]
if !ok {
return ErrIndexDoesNotExist
}
fld, ok := idx.Fields[field]
if !ok {
return ErrFieldDoesNotExist
}
delete(fld.Views, view)
return nil
}
var InMemSharder Sharder = &inMemSharder{
shards: make(map[string][]byte),
}
type inMemSharder struct {
mu sync.RWMutex
shards map[string][]byte
}
func (s *inMemSharder) Shards(ctx context.Context, index, field string) ([][]byte, error) {
key := path.Join("/shard/", index, field)
s.mu.RLock()
defer s.mu.RUnlock()
b := s.shards[key]
if b == nil {
return nil, nil
}
return [][]byte{b}, nil
}
func (s *inMemSharder) SetShards(ctx context.Context, index, field string, shards []byte) error {
key := path.Join("/shard/", index, field)
s.mu.Lock()
defer s.mu.Unlock()
s.shards[key] = make([]byte, len(shards))
copy(s.shards[key], shards)
return nil
}