zitadel/internal/eventstore/v3/push.go
Silvan 9c3e5e467b
perf(query): remove transactions for queries (#8614)
# Which Problems Are Solved

Queries currently execute 3 statements, begin, query, commit

# How the Problems Are Solved

remove transaction handling from query methods in database package

# Additional Changes

- Bump versions of `core_grpc_dependencies`-receipt in Makefile

# Additional info

During load tests we saw a lot of idle transactions of `zitadel_queries`
application name which is the connection pool used to query data in
zitadel. Executed query:

`select query_start - xact_start, pid, application_name, backend_start,
xact_start, query_start, state_change, wait_event_type,
wait_event,substring(query, 1, 200) query from pg_stat_activity where
datname = 'zitadel' and state <> 'idle';`

Mostly the last query executed was `begin isolation level read committed
read only`.

example: 

```
    ?column?     |  pid  |      application_name      |         backend_start         |          xact_start           |          query_start          |         state_change          | wait_event_type |  wait_event  |                                                                                                  query                                                                                                   
-----------------+-------+----------------------------+-------------------------------+-------------------------------+-------------------------------+-------------------------------+-----------------+--------------+----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------
 00:00:00        | 33030 | zitadel_queries            | 2024-10-16 16:25:53.906036+00 | 2024-10-16 16:30:19.191661+00 | 2024-10-16 16:30:19.191661+00 | 2024-10-16 16:30:19.19169+00  | Client          | ClientRead   | begin isolation level read committed read only
 00:00:00        | 33035 | zitadel_queries            | 2024-10-16 16:25:53.909629+00 | 2024-10-16 16:30:19.19179+00  | 2024-10-16 16:30:19.19179+00  | 2024-10-16 16:30:19.191805+00 | Client          | ClientRead   | begin isolation level read committed read only
 00:00:00.00412  | 33028 | zitadel_queries            | 2024-10-16 16:25:53.904247+00 | 2024-10-16 16:30:19.187734+00 | 2024-10-16 16:30:19.191854+00 | 2024-10-16 16:30:19.191964+00 | Client          | ClientRead   | SELECT created_at, event_type, "sequence", "position", payload, creator, "owner", instance_id, aggregate_type, aggregate_id, revision FROM eventstore.events2 WHERE instance_id = $1 AND aggregate_type 
 00:00:00.084662 | 33134 | zitadel_es_pusher          | 2024-10-16 16:29:54.979692+00 | 2024-10-16 16:30:19.178578+00 | 2024-10-16 16:30:19.26324+00  | 2024-10-16 16:30:19.263267+00 | Client          | ClientRead   | RELEASE SAVEPOINT cockroach_restart
 00:00:00.084768 | 33139 | zitadel_es_pusher          | 2024-10-16 16:29:54.979585+00 | 2024-10-16 16:30:19.180762+00 | 2024-10-16 16:30:19.26553+00  | 2024-10-16 16:30:19.265531+00 | LWLock          | WALWriteLock | commit
 00:00:00.077377 | 33136 | zitadel_es_pusher          | 2024-10-16 16:29:54.978582+00 | 2024-10-16 16:30:19.187883+00 | 2024-10-16 16:30:19.26526+00  | 2024-10-16 16:30:19.265431+00 | Client          | ClientRead   | WITH existing AS (                                                                                                                                                                                      +
                 |       |                            |                               |                               |                               |                               |                 |              |     (SELECT instance_id, aggregate_type, aggregate_id, "sequence" FROM eventstore.events2 WHERE instance_id = $1 AND aggregate_type = $2 AND aggregate_id = $3 ORDER BY "sequence" DE
 00:00:00.012309 | 33123 | zitadel_es_pusher          | 2024-10-16 16:29:54.963484+00 | 2024-10-16 16:30:19.175066+00 | 2024-10-16 16:30:19.187375+00 | 2024-10-16 16:30:19.187376+00 | IO              | WalSync      | commit
 00:00:00        | 33034 | zitadel_queries            | 2024-10-16 16:25:53.90791+00  | 2024-10-16 16:30:19.262921+00 | 2024-10-16 16:30:19.262921+00 | 2024-10-16 16:30:19.263133+00 | Client          | ClientRead   | begin isolation level read committed read only
 00:00:00        | 33039 | zitadel_queries            | 2024-10-16 16:25:53.914106+00 | 2024-10-16 16:30:19.191676+00 | 2024-10-16 16:30:19.191676+00 | 2024-10-16 16:30:19.191687+00 | Client          | ClientRead   | begin isolation level read committed read only
 00:00:00.24539  | 33083 | zitadel_projection_spooler | 2024-10-16 16:27:49.895548+00 | 2024-10-16 16:30:19.020058+00 | 2024-10-16 16:30:19.265448+00 | 2024-10-16 16:30:19.26546+00  | Client          | ClientRead   | SAVEPOINT exec_stmt
 00:00:00        | 33125 | zitadel_es_pusher          | 2024-10-16 16:29:54.963859+00 | 2024-10-16 16:30:19.191715+00 | 2024-10-16 16:30:19.191715+00 | 2024-10-16 16:30:19.191729+00 | Client          | ClientRead   | begin
 00:00:00.004292 | 33032 | zitadel_queries            | 2024-10-16 16:25:53.906624+00 | 2024-10-16 16:30:19.187713+00 | 2024-10-16 16:30:19.192005+00 | 2024-10-16 16:30:19.192062+00 | Client          | ClientRead   | SELECT created_at, event_type, "sequence", "position", payload, creator, "owner", instance_id, aggregate_type, aggregate_id, revision FROM eventstore.events2 WHERE instance_id = $1 AND aggregate_type 
 00:00:00        | 33031 | zitadel_queries            | 2024-10-16 16:25:53.906422+00 | 2024-10-16 16:30:19.191625+00 | 2024-10-16 16:30:19.191625+00 | 2024-10-16 16:30:19.191645+00 | Client          | ClientRead   | begin isolation level read committed read only

```

The amount of idle transactions is significantly less if the query
transactions are removed:

example: 

```
    ?column?     |  pid  |      application_name      |         backend_start         |          xact_start           |          query_start          |         state_change          | wait_event_type | wait_event |                                                                                                  query                                                                                                   
-----------------+-------+----------------------------+-------------------------------+-------------------------------+-------------------------------+-------------------------------+-----------------+------------+----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------
 00:00:00.000094 | 32741 | zitadel_queries            | 2024-10-16 16:23:49.73935+00  | 2024-10-16 16:24:59.785589+00 | 2024-10-16 16:24:59.785683+00 | 2024-10-16 16:24:59.785684+00 |                 |            | SELECT created_at, event_type, "sequence", "position", payload, creator, "owner", instance_id, aggregate_type, aggregate_id, revision FROM eventstore.events2 WHERE instance_id = $1 AND aggregate_type 
 00:00:00        | 32762 | zitadel_es_pusher          | 2024-10-16 16:24:02.275136+00 | 2024-10-16 16:24:59.784586+00 | 2024-10-16 16:24:59.784586+00 | 2024-10-16 16:24:59.784607+00 | Client          | ClientRead | begin
 00:00:00.000167 | 32742 | zitadel_queries            | 2024-10-16 16:23:49.740489+00 | 2024-10-16 16:24:59.784274+00 | 2024-10-16 16:24:59.784441+00 | 2024-10-16 16:24:59.784442+00 |                 |            | with usr as (                                                                                                                                                                                           +
                 |       |                            |                               |                               |                               |                               |                 |            |         select u.id, u.creation_date, u.change_date, u.sequence, u.state, u.resource_owner, u.username, n.login_name as preferred_login_name                                                            +
                 |       |                            |                               |                               |                               |                               |                 |            |         from projections.users13 u                                                                                                                                                                      +
                 |       |                            |                               |                               |                               |                               |                 |            |         left join projections.l
 00:00:00.256014 | 32759 | zitadel_projection_spooler | 2024-10-16 16:24:01.418429+00 | 2024-10-16 16:24:59.52959+00  | 2024-10-16 16:24:59.785604+00 | 2024-10-16 16:24:59.785649+00 | Client          | ClientRead | UPDATE projections.milestones SET reached_date = $1 WHERE (instance_id = $2) AND (type = $3) AND (reached_date IS NULL)
 00:00:00.014199 | 32773 | zitadel_es_pusher          | 2024-10-16 16:24:02.320404+00 | 2024-10-16 16:24:59.769509+00 | 2024-10-16 16:24:59.783708+00 | 2024-10-16 16:24:59.783709+00 | IO              | WalSync    | commit
 00:00:00        | 32765 | zitadel_es_pusher          | 2024-10-16 16:24:02.28173+00  | 2024-10-16 16:24:59.780413+00 | 2024-10-16 16:24:59.780413+00 | 2024-10-16 16:24:59.780426+00 | Client          | ClientRead | begin
 00:00:00.012729 | 32777 | zitadel_es_pusher          | 2024-10-16 16:24:02.339737+00 | 2024-10-16 16:24:59.767432+00 | 2024-10-16 16:24:59.780161+00 | 2024-10-16 16:24:59.780195+00 | Client          | ClientRead | RELEASE SAVEPOINT cockroach_restart
```

---------

Co-authored-by: Tim Möhlmann <tim+github@zitadel.com>
Co-authored-by: Livio Spring <livio.a@gmail.com>
Co-authored-by: Max Peintner <max@caos.ch>
Co-authored-by: Elio Bischof <elio@zitadel.com>
Co-authored-by: Stefan Benz <46600784+stebenz@users.noreply.github.com>
Co-authored-by: Miguel Cabrerizo <30386061+doncicuto@users.noreply.github.com>
Co-authored-by: Joakim Lodén <Loddan@users.noreply.github.com>
Co-authored-by: Yxnt <Yxnt@users.noreply.github.com>
Co-authored-by: Stefan Benz <stefan@caos.ch>
Co-authored-by: Harsha Reddy <harsha.reddy@klaviyo.com>
Co-authored-by: Zach H <zhirschtritt@gmail.com>
2024-11-04 10:06:14 +01:00

188 lines
4.8 KiB
Go

package eventstore
import (
"context"
"database/sql"
_ "embed"
"errors"
"fmt"
"strconv"
"strings"
"github.com/cockroachdb/cockroach-go/v2/crdb"
"github.com/jackc/pgx/v5/pgconn"
"github.com/zitadel/logging"
"github.com/zitadel/zitadel/internal/eventstore"
"github.com/zitadel/zitadel/internal/telemetry/tracing"
"github.com/zitadel/zitadel/internal/zerrors"
)
func (es *Eventstore) Push(ctx context.Context, commands ...eventstore.Command) (events []eventstore.Event, err error) {
ctx, span := tracing.NewSpan(ctx)
defer func() { span.EndWithError(err) }()
tx, err := es.client.BeginTx(ctx, nil)
if err != nil {
return nil, err
}
// tx is not closed because [crdb.ExecuteInTx] takes care of that
var (
sequences []*latestSequence
)
err = crdb.ExecuteInTx(ctx, &transaction{tx}, func() (err error) {
inTxCtx, span := tracing.NewSpan(ctx)
defer func() { span.EndWithError(err) }()
sequences, err = latestSequences(inTxCtx, tx, commands)
if err != nil {
return err
}
events, err = insertEvents(inTxCtx, tx, sequences, commands)
if err != nil {
return err
}
if err = handleUniqueConstraints(inTxCtx, tx, commands); err != nil {
return err
}
// CockroachDB by default does not allow multiple modifications of the same table using ON CONFLICT
// Thats why we enable it manually
if es.client.Type() == "cockroach" {
_, err = tx.Exec("SET enable_multiple_modifications_of_table = on")
if err != nil {
return err
}
}
return handleFieldCommands(inTxCtx, tx, commands)
})
if err != nil {
return nil, err
}
return events, nil
}
//go:embed push.sql
var pushStmt string
func insertEvents(ctx context.Context, tx *sql.Tx, sequences []*latestSequence, commands []eventstore.Command) ([]eventstore.Event, error) {
events, placeholders, args, err := mapCommands(commands, sequences)
if err != nil {
return nil, err
}
rows, err := tx.QueryContext(ctx, fmt.Sprintf(pushStmt, strings.Join(placeholders, ", ")), args...)
if err != nil {
return nil, err
}
defer rows.Close()
for i := 0; rows.Next(); i++ {
err = rows.Scan(&events[i].(*event).createdAt, &events[i].(*event).position)
if err != nil {
logging.WithError(err).Warn("failed to scan events")
return nil, err
}
}
if err := rows.Err(); err != nil {
pgErr := new(pgconn.PgError)
if errors.As(err, &pgErr) {
// Check if push tries to write an event just written
// by another transaction
if pgErr.Code == "40001" {
// TODO: @livio-a should we return the parent or not?
return nil, zerrors.ThrowInvalidArgument(err, "V3-p5xAn", "Errors.AlreadyExists")
}
}
logging.WithError(rows.Err()).Warn("failed to push events")
return nil, zerrors.ThrowInternal(err, "V3-VGnZY", "Errors.Internal")
}
return events, nil
}
const argsPerCommand = 10
func mapCommands(commands []eventstore.Command, sequences []*latestSequence) (events []eventstore.Event, placeholders []string, args []any, err error) {
events = make([]eventstore.Event, len(commands))
args = make([]any, 0, len(commands)*argsPerCommand)
placeholders = make([]string, len(commands))
for i, command := range commands {
sequence := searchSequenceByCommand(sequences, command)
if sequence == nil {
logging.WithFields(
"aggType", command.Aggregate().Type,
"aggID", command.Aggregate().ID,
"instance", command.Aggregate().InstanceID,
).Panic("no sequence found")
// added return for linting
return nil, nil, nil, nil
}
sequence.sequence++
events[i], err = commandToEvent(sequence, command)
if err != nil {
return nil, nil, nil, err
}
placeholders[i] = fmt.Sprintf(pushPlaceholderFmt,
i*argsPerCommand+1,
i*argsPerCommand+2,
i*argsPerCommand+3,
i*argsPerCommand+4,
i*argsPerCommand+5,
i*argsPerCommand+6,
i*argsPerCommand+7,
i*argsPerCommand+8,
i*argsPerCommand+9,
i*argsPerCommand+10,
)
revision, err := strconv.Atoi(strings.TrimPrefix(string(events[i].(*event).aggregate.Version), "v"))
if err != nil {
return nil, nil, nil, zerrors.ThrowInternal(err, "V3-JoZEp", "Errors.Internal")
}
args = append(args,
events[i].(*event).aggregate.InstanceID,
events[i].(*event).aggregate.ResourceOwner,
events[i].(*event).aggregate.Type,
events[i].(*event).aggregate.ID,
revision,
events[i].(*event).creator,
events[i].(*event).typ,
events[i].(*event).payload,
events[i].(*event).sequence,
i,
)
}
return events, placeholders, args, nil
}
type transaction struct {
*sql.Tx
}
var _ crdb.Tx = (*transaction)(nil)
func (t *transaction) Exec(ctx context.Context, query string, args ...interface{}) error {
_, err := t.Tx.ExecContext(ctx, query, args...)
return err
}
func (t *transaction) Commit(ctx context.Context) error {
return t.Tx.Commit()
}
func (t *transaction) Rollback(ctx context.Context) error {
return t.Tx.Rollback()
}