feat: miner deps: harmonydb

This commit is contained in:
Andrew Jackson (Ajax)
2023-07-14 18:05:49 -05:00
parent 225cf0fa7f
commit d17bfb14f8
28 changed files with 993 additions and 89 deletions
+35
View File
@@ -0,0 +1,35 @@
/*
# HarmonyDB provides database abstractions over SP-wide Postgres-compatible instance(s).
# Features
Rolling to secondary database servers on connection failure
Convenience features for Go + SQL
Prevention of SQL injection vulnerabilities
Monitors behavior via Prometheus stats and logging of errors.
# Usage
Processes should use New() to instantiate a *DB and keep it.
Consumers can use this *DB concurrently.
Creating and changing tables & views should happen in ./sql/ folder.
Name the file "today's date" in the format: YYYYMMDD.sql (ex: 20231231.sql for the year's last day)
a. CREATE TABLE should NOT have a schema:
GOOD: CREATE TABLE foo ();
BAD: CREATE TABLE me.foo ();
b. Schema is managed for you. It provides isolation for integraton tests & multi-use.
c. Git Merges: All run once, so old-after-new is OK when there are no deps.
d. NEVER change shipped sql files. Have later files make corrections.
e. Anything not ran will be ran, so an older date making it to master is OK.
Write SQL with context, raw strings, and args:
name := "Alice"
var ID int
err := QueryRow(ctx, "SELECT id from people where first_name=?", name).Scan(&ID)
fmt.Println(ID)
Note: Scan() is column-oriented, while Select() & StructScan() is field name/tag oriented.
*/
package harmonydb
+266
View File
@@ -0,0 +1,266 @@
package harmonydb
import (
"context"
"embed"
"errors"
"fmt"
"math/rand"
"regexp"
"sort"
"strconv"
"strings"
"time"
logging "github.com/ipfs/go-log/v2"
"github.com/jackc/pgx/v5"
"github.com/jackc/pgx/v5/pgconn"
"github.com/jackc/pgx/v5/pgxpool"
"github.com/filecoin-project/lotus/node/config"
)
type ITestID string
// ItestNewID see ITestWithID doc
func ITestNewID() ITestID {
return ITestID(strconv.Itoa(rand.Intn(99999)))
}
type DB struct {
pgx *pgxpool.Pool
cfg *pgxpool.Config
schema string
hostnames []string
log func(string)
}
var logger = logging.Logger("harmonydb")
// NewFromConfig is a convenience function.
// In usage:
//
// db, err := NewFromConfig(config.HarmonyDB) // in binary init
func NewFromConfig(cfg config.HarmonyDB) (*DB, error) {
return New(
cfg.Hosts,
cfg.Username,
cfg.Password,
cfg.Database,
cfg.Port,
"",
func(s string) { logger.Error(s) },
)
}
func NewFromConfigWithITestID(cfg config.HarmonyDB) func(id ITestID) (*DB, error) {
return func(id ITestID) (*DB, error) {
return New(
cfg.Hosts,
cfg.Username,
cfg.Password,
cfg.Database,
cfg.Port,
id,
func(s string) { logger.Error(s) },
)
}
}
// New is to be called once per binary to establish the pool.
// log() is for errors. It returns an upgraded database's connection.
// This entry point serves both production and integration tests, so it's more DI.
func New(hosts []string, username, password, database, port string, itestID ITestID, log func(string)) (*DB, error) {
itest := string(itestID)
connString := ""
if len(hosts) > 0 {
connString = "host=" + hosts[0] + " "
}
for k, v := range map[string]string{"user": username, "password": password, "dbname": database, "port": port} {
if strings.TrimSpace(v) != "" {
connString += k + "=" + v + " "
}
}
schema := "lotus"
if itest != "" {
schema = "itest_" + itest
}
if err := ensureSchemaExists(connString, schema); err != nil {
return nil, err
}
cfg, err := pgxpool.ParseConfig(connString + "search_path=" + schema)
if err != nil {
return nil, err
}
// enable multiple fallback hosts.
for _, h := range hosts[1:] {
cfg.ConnConfig.Fallbacks = append(cfg.ConnConfig.Fallbacks, &pgconn.FallbackConfig{Host: h})
}
cfg.ConnConfig.OnNotice = func(conn *pgconn.PgConn, n *pgconn.Notice) {
log("database notice: " + n.Message + ": " + n.Detail)
DBMeasures.Errors.M(1)
}
db := DB{cfg: cfg, schema: schema, hostnames: hosts, log: log} // pgx populated in AddStatsAndConnect
if err := db.addStatsAndConnect(); err != nil {
return nil, err
}
return &db, db.upgrade()
}
type tracer struct {
}
type ctxkey string
var sqlStart = ctxkey("sqlStart")
func (t tracer) TraceQueryStart(ctx context.Context, conn *pgx.Conn, data pgx.TraceQueryStartData) context.Context {
return context.WithValue(ctx, sqlStart, time.Now())
}
func (t tracer) TraceQueryEnd(ctx context.Context, conn *pgx.Conn, data pgx.TraceQueryEndData) {
DBMeasures.Hits.M(1)
ms := time.Since(ctx.Value(sqlStart).(time.Time)).Milliseconds()
DBMeasures.TotalWait.M(ms)
DBMeasures.Waits.Observe(float64(ms))
if data.Err != nil {
DBMeasures.Errors.M(1)
}
// Can log what type of query it is, but not what tables
// Can log rows affected.
}
// addStatsAndConnect connects a prometheus logger. Be sure to run this before using the DB.
func (db *DB) addStatsAndConnect() error {
db.cfg.ConnConfig.Tracer = tracer{}
hostnameToIndex := map[string]float64{}
for i, h := range db.hostnames {
hostnameToIndex[h] = float64(i)
}
db.cfg.AfterConnect = func(ctx context.Context, c *pgx.Conn) error {
s := db.pgx.Stat()
DBMeasures.OpenConnections.M(int64(s.TotalConns()))
DBMeasures.WhichHost.Observe(hostnameToIndex[c.Config().Host])
//FUTURE place for any connection seasoning
return nil
}
var err error
db.pgx, err = pgxpool.NewWithConfig(context.Background(), db.cfg)
if err != nil {
db.log(fmt.Sprintf("Unable to connect to database: %v\n", err))
return err
}
return nil
}
// ITestDeleteAll will delete everything created for "this" integration test.
// This must be called at the end of each integration test.
func (db *DB) ITestDeleteAll() {
if !strings.HasPrefix(db.schema, "itest_") {
fmt.Println("Warning: this should never be called on anything but an itest schema.")
return
}
defer db.pgx.Close()
_, err := db.pgx.Exec(context.Background(), "DROP SCHEMA "+db.schema+" CASCADE")
if err != nil {
fmt.Println("warning: unclean itest shutdown: cannot delete schema: " + err.Error())
return
}
}
var schemaREString = "^[A-Za-z0-9_]+$"
var schemaRE = regexp.MustCompile(schemaREString)
func ensureSchemaExists(connString, schema string) error {
// FUTURE allow using fallback DBs for start-up.
p, err := pgx.Connect(context.Background(), connString)
if err != nil {
return err
}
defer func() { _ = p.Close(context.Background()) }()
if len(schema) < 5 || !schemaRE.MatchString(schema) {
return errors.New("schema must be of the form " + schemaREString + "\n Got: " + schema)
}
_, err = p.Exec(context.Background(), "CREATE SCHEMA IF NOT EXISTS "+schema)
if err != nil {
return fmt.Errorf("cannot create schema: %w", err)
}
return nil
}
//go:embed sql
var fs embed.FS
func (db *DB) upgrade() error {
// Does the version table exist? if not, make it.
// NOTE: This cannot change except via the next sql file.
err := db.Exec(context.Background(), `CREATE TABLE IF NOT EXISTS base (
id SERIAL PRIMARY KEY,
entry CHAR(12),
applied TIMESTAMP DEFAULT current_timestamp
)`)
if err != nil {
db.log("Upgrade failed.")
return err
}
// __Run scripts in order.__
landed := map[string]bool{}
{
var landedEntries []struct{ Entry string }
err = db.Select(context.Background(), &landedEntries, "SELECT entry FROM base")
if err != nil {
db.log("Cannot read entries: " + err.Error())
return err
}
for _, l := range landedEntries {
landed[l.Entry] = true
}
}
dir, err := fs.ReadDir("sql")
if err != nil {
db.log("Cannot read fs entries: " + err.Error())
return err
}
sort.Slice(dir, func(i, j int) bool { return dir[i].Name() < dir[j].Name() })
for _, e := range dir {
name := e.Name()
if landed[name] || !strings.HasSuffix(name, ".sql") {
continue
}
file, err := fs.ReadFile("sql/" + name)
if err != nil {
db.log("weird embed file read err")
return err
}
for _, s := range strings.Split(string(file), ";") { // Implement the changes.
if len(strings.TrimSpace(s)) == 0 {
continue
}
_, err = db.pgx.Exec(context.Background(), s)
if err != nil {
db.log(fmt.Sprintf("Could not upgrade! File %s, Query: %s, Returned: %s", name, s, err.Error()))
return err
}
}
// Mark Completed.
err = db.Exec(context.Background(), "INSERT INTO base (entry) VALUES ($1)", name)
if err != nil {
db.log("Cannot update base: " + err.Error())
return fmt.Errorf("cannot insert into base: %w", err)
}
}
return nil
}
+77
View File
@@ -0,0 +1,77 @@
package harmonydb
import (
"github.com/prometheus/client_golang/prometheus"
"go.opencensus.io/stats"
"go.opencensus.io/stats/view"
"go.opencensus.io/tag"
"github.com/filecoin-project/lotus/metrics"
)
var (
dbTag, _ = tag.NewKey("db_name")
pre = "harmonydb_base_"
waitsBuckets = []float64{0, 10, 20, 30, 50, 80, 130, 210, 340, 550, 890}
whichHostBuckets = []float64{0, 1, 2, 3, 4, 5}
)
// DBMeasures groups all db metrics.
var DBMeasures = struct {
Hits *stats.Int64Measure
TotalWait *stats.Int64Measure
Waits prometheus.Histogram
OpenConnections *stats.Int64Measure
Errors *stats.Int64Measure
WhichHost prometheus.Histogram
}{
Hits: stats.Int64(pre+"hits", "Total number of uses.", stats.UnitDimensionless),
TotalWait: stats.Int64(pre+"total_wait", "Total delay. A numerator over hits to get average wait.", stats.UnitMilliseconds),
Waits: prometheus.NewHistogram(prometheus.HistogramOpts{
Name: pre + "waits",
Buckets: waitsBuckets,
Help: "The histogram of waits for query completions.",
}),
OpenConnections: stats.Int64(pre+"open_connections", "Total connection count.", stats.UnitDimensionless),
Errors: stats.Int64(pre+"errors", "Total error count.", stats.UnitDimensionless),
WhichHost: prometheus.NewHistogram(prometheus.HistogramOpts{
Name: pre + "which_host",
Buckets: whichHostBuckets,
Help: "The index of the hostname being used",
}),
}
// CacheViews groups all cache-related default views.
func init() {
metrics.RegisterViews(
&view.View{
Measure: DBMeasures.Hits,
Aggregation: view.Sum(),
TagKeys: []tag.Key{dbTag},
},
&view.View{
Measure: DBMeasures.TotalWait,
Aggregation: view.Sum(),
TagKeys: []tag.Key{dbTag},
},
&view.View{
Measure: DBMeasures.OpenConnections,
Aggregation: view.LastValue(),
TagKeys: []tag.Key{dbTag},
},
&view.View{
Measure: DBMeasures.Errors,
Aggregation: view.Sum(),
TagKeys: []tag.Key{dbTag},
},
)
err := prometheus.Register(DBMeasures.Waits)
if err != nil {
panic(err)
}
err = prometheus.Register(DBMeasures.WhichHost)
if err != nil {
panic(err)
}
}
+6
View File
@@ -0,0 +1,6 @@
CREATE TABLE itest_scratch (
id SERIAL PRIMARY KEY,
content TEXT,
some_int INTEGER,
update_time TIMESTAMP DEFAULT current_timestamp
)
+149
View File
@@ -0,0 +1,149 @@
package harmonydb
import (
"context"
"github.com/georgysavva/scany/v2/pgxscan"
"github.com/jackc/pgx/v5"
)
type Intf interface {
ITestDeleteAll()
Exec(ctx context.Context, sql rawStringOnly, arguments ...any) error
Query(ctx context.Context, sql rawStringOnly, arguments ...any) (*Query, error)
QueryRow(ctx context.Context, sql rawStringOnly, arguments ...any) Row
Select(ctx context.Context, sliceOfStructPtr any, sql rawStringOnly, arguments ...any) error
BeginTransaction(ctx context.Context, f func(t *Transaction) (commit bool)) (retErr error)
}
// rawStringOnly is _intentionally_private_ to force only basic strings in SQL queries.
// In any package, raw strings will satisfy compilation. Ex:
//
// harmonydb.Exec("INSERT INTO version (number) VALUES (1)")
//
// This prevents SQL injection attacks where the input contains query fragments.
type rawStringOnly string
// Exec executes changes (INSERT, DELETE, or UPDATE).
// Note, for CREATE & DROP please keep these permanent and express
// them in the ./sql/ files (next number).
func (db *DB) Exec(ctx context.Context, sql rawStringOnly, arguments ...any) error {
_, err := db.pgx.Exec(ctx, string(sql), arguments...)
return err
}
type Qry interface {
Next() bool
Err() error
Close()
Scan(...any) error
Values() ([]any, error)
}
// Query offers Next/Err/Close/Scan/Values/StructScan
type Query struct {
Qry
}
// Query allows iterating returned values to save memory consumption
// with the downside of needing to `defer q.Close()`. For a simpler interface,
// try Select()
// Next() must be called to advance the row cursor, including the first time:
// Ex:
// q, err := db.Query(ctx, "SELECT id, name FROM users")
// handleError(err)
// defer q.Close()
//
// for q.Next() {
// var id int
// var name string
// handleError(q.Scan(&id, &name))
// fmt.Println(id, name)
// }
func (db *DB) Query(ctx context.Context, sql rawStringOnly, arguments ...any) (*Query, error) {
q, err := db.pgx.Query(ctx, string(sql), arguments...)
return &Query{q}, err
}
func (q *Query) StructScan(s any) error {
return pgxscan.ScanRow(s, q.Qry.(pgx.Rows))
}
type Row interface {
Scan(...any) error
}
// QueryRow gets 1 row using column order matching.
// This is a timesaver for the special case of wanting the first row returned only.
// EX:
//
// var name, pet string
// var ID = 123
// err := db.QueryRow(ctx, "SELECT name, pet FROM users WHERE ID=?", ID).Scan(&name, &pet)
func (db *DB) QueryRow(ctx context.Context, sql rawStringOnly, arguments ...any) Row {
return db.pgx.QueryRow(ctx, string(sql), arguments...)
}
/*
Select multiple rows into a slice using name matching
Ex:
type user struct {
Name string
ID int
Number string `db:"tel_no"`
}
var users []user
pet := "cat"
err := db.Select(ctx, &users, "SELECT name, id, tel_no FROM customers WHERE pet=?", pet)
*/
func (db *DB) Select(ctx context.Context, sliceOfStructPtr any, sql rawStringOnly, arguments ...any) error {
return pgxscan.Select(ctx, db.pgx, sliceOfStructPtr, string(sql), arguments...)
}
type Transaction struct {
pgx.Tx
}
// BeginTransaction is how you can access transactions using this library.
// The entire transaction happens in the function passed in.
// The return must be true or a rollback will occur.
func (db *DB) BeginTransaction(ctx context.Context, f func(t *Transaction) (commit bool)) (retErr error) {
tx, err := db.pgx.BeginTx(ctx, pgx.TxOptions{})
if err != nil {
return err
}
var commit bool
defer func() { // Panic clean-up.
if !commit {
retErr = tx.Rollback(ctx)
}
}()
commit = f(&Transaction{tx})
if commit {
return tx.Commit(ctx)
}
return nil
}
// Exec in a transaction.
func (t *Transaction) Exec(ctx context.Context, sql rawStringOnly, arguments ...any) error {
_, err := t.Tx.Exec(ctx, string(sql), arguments...)
return err
}
// Query in a transaction.
func (t *Transaction) Query(ctx context.Context, sql rawStringOnly, arguments ...any) (*Query, error) {
q, err := t.Tx.Query(ctx, string(sql), arguments...)
return &Query{q}, err
}
// QueryRow in a transaction.
func (t *Transaction) QueryRow(ctx context.Context, sql rawStringOnly, arguments ...any) Row {
return t.Tx.QueryRow(ctx, string(sql), arguments...)
}
// Select in a transaction.
func (t *Transaction) Select(ctx context.Context, sliceOfStructPtr any, sql rawStringOnly, arguments ...any) error {
return pgxscan.Select(ctx, t.Tx, sliceOfStructPtr, string(sql), arguments...)
}