-
Notifications
You must be signed in to change notification settings - Fork 1
/
copy_from.go
249 lines (208 loc) · 6.36 KB
/
copy_from.go
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
package pgx
import (
"bytes"
"context"
"fmt"
"io"
"github.com/jackc/pgx/v5/internal/pgio"
"github.com/jackc/pgx/v5/pgconn"
)
// CopyFromRows returns a CopyFromSource interface over the provided rows slice
// making it usable by *Conn.CopyFrom.
func CopyFromRows(rows [][]any) CopyFromSource {
return ©FromRows{rows: rows, idx: -1}
}
type copyFromRows struct {
rows [][]any
idx int
}
func (ctr *copyFromRows) Next() bool {
ctr.idx++
return ctr.idx < len(ctr.rows)
}
func (ctr *copyFromRows) Values() ([]any, error) {
return ctr.rows[ctr.idx], nil
}
func (ctr *copyFromRows) Err() error {
return nil
}
// CopyFromSlice returns a CopyFromSource interface over a dynamic func
// making it usable by *Conn.CopyFrom.
func CopyFromSlice(length int, next func(int) ([]any, error)) CopyFromSource {
return ©FromSlice{next: next, idx: -1, len: length}
}
type copyFromSlice struct {
next func(int) ([]any, error)
idx int
len int
err error
}
func (cts *copyFromSlice) Next() bool {
cts.idx++
return cts.idx < cts.len
}
func (cts *copyFromSlice) Values() ([]any, error) {
values, err := cts.next(cts.idx)
if err != nil {
cts.err = err
}
return values, err
}
func (cts *copyFromSlice) Err() error {
return cts.err
}
// CopyFromSource is the interface used by *Conn.CopyFrom as the source for copy data.
type CopyFromSource interface {
// Next returns true if there is another row and makes the next row data
// available to Values(). When there are no more rows available or an error
// has occurred it returns false.
Next() bool
// Values returns the values for the current row.
Values() ([]any, error)
// Err returns any error that has been encountered by the CopyFromSource. If
// this is not nil *Conn.CopyFrom will abort the copy.
Err() error
}
type copyFrom struct {
conn *Conn
tableName Identifier
columnNames []string
rowSrc CopyFromSource
readerErrChan chan error
mode QueryExecMode
}
func (ct *copyFrom) run(ctx context.Context) (int64, error) {
if ct.conn.copyFromTracer != nil {
ctx = ct.conn.copyFromTracer.TraceCopyFromStart(ctx, ct.conn, TraceCopyFromStartData{
TableName: ct.tableName,
ColumnNames: ct.columnNames,
})
}
quotedTableName := ct.tableName.Sanitize()
cbuf := &bytes.Buffer{}
for i, cn := range ct.columnNames {
if i != 0 {
cbuf.WriteString(", ")
}
cbuf.WriteString(quoteIdentifier(cn))
}
quotedColumnNames := cbuf.String()
var sd *pgconn.StatementDescription
switch ct.mode {
case QueryExecModeExec, QueryExecModeSimpleProtocol:
// These modes don't support the binary format. Before the inclusion of the
// QueryExecModes, Conn.Prepare was called on every COPY operation to get
// the OIDs. These prepared statements were not cached.
//
// Since that's the same behavior provided by QueryExecModeDescribeExec,
// we'll default to that mode.
ct.mode = QueryExecModeDescribeExec
fallthrough
case QueryExecModeCacheStatement, QueryExecModeCacheDescribe, QueryExecModeDescribeExec:
var err error
sd, err = ct.conn.getStatementDescription(
ctx,
ct.mode,
fmt.Sprintf("select %s from %s", quotedColumnNames, quotedTableName),
)
if err != nil {
return 0, fmt.Errorf("statement description failed: %w", err)
}
default:
return 0, fmt.Errorf("unknown QueryExecMode: %v", ct.mode)
}
r, w := io.Pipe()
doneChan := make(chan struct{})
go func() {
defer close(doneChan)
// Purposely NOT using defer w.Close(). See https://github.com/golang/go/issues/24283.
buf := ct.conn.wbuf
buf = append(buf, "PGCOPY\n\377\r\n\000"...)
buf = pgio.AppendInt32(buf, 0)
buf = pgio.AppendInt32(buf, 0)
moreRows := true
for moreRows {
var err error
moreRows, buf, err = ct.buildCopyBuf(buf, sd)
if err != nil {
w.CloseWithError(err)
return
}
if ct.rowSrc.Err() != nil {
w.CloseWithError(ct.rowSrc.Err())
return
}
if len(buf) > 0 {
_, err = w.Write(buf)
if err != nil {
w.Close()
return
}
}
buf = buf[:0]
}
w.Close()
}()
commandTag, err := ct.conn.pgConn.CopyFrom(ctx, r, fmt.Sprintf("copy %s ( %s ) from stdin binary;", quotedTableName, quotedColumnNames))
r.Close()
<-doneChan
if ct.conn.copyFromTracer != nil {
ct.conn.copyFromTracer.TraceCopyFromEnd(ctx, ct.conn, TraceCopyFromEndData{
CommandTag: commandTag,
Err: err,
})
}
return commandTag.RowsAffected(), err
}
func (ct *copyFrom) buildCopyBuf(buf []byte, sd *pgconn.StatementDescription) (bool, []byte, error) {
const sendBufSize = 65536 - 5 // The packet has a 5-byte header
lastBufLen := 0
largestRowLen := 0
for ct.rowSrc.Next() {
lastBufLen = len(buf)
values, err := ct.rowSrc.Values()
if err != nil {
return false, nil, err
}
if len(values) != len(ct.columnNames) {
return false, nil, fmt.Errorf("expected %d values, got %d values", len(ct.columnNames), len(values))
}
buf = pgio.AppendInt16(buf, int16(len(ct.columnNames)))
for i, val := range values {
buf, err = encodeCopyValue(ct.conn.typeMap, buf, sd.Fields[i].DataTypeOID, val)
if err != nil {
return false, nil, err
}
}
rowLen := len(buf) - lastBufLen
if rowLen > largestRowLen {
largestRowLen = rowLen
}
// Try not to overflow size of the buffer PgConn.CopyFrom will be reading into. If that happens then the nature of
// io.Pipe means that the next Read will be short. This can lead to pathological send sizes such as 65531, 13, 65531
// 13, 65531, 13, 65531, 13.
if len(buf) > sendBufSize-largestRowLen {
return true, buf, nil
}
}
return false, buf, nil
}
// CopyFrom uses the PostgreSQL copy protocol to perform bulk data insertion. It returns the number of rows copied and
// an error.
//
// CopyFrom requires all values use the binary format. A pgtype.Type that supports the binary format must be registered
// for the type of each column. Almost all types implemented by pgx support the binary format.
//
// Even though enum types appear to be strings they still must be registered to use with CopyFrom. This can be done with
// Conn.LoadType and pgtype.Map.RegisterType.
func (c *Conn) CopyFrom(ctx context.Context, tableName Identifier, columnNames []string, rowSrc CopyFromSource) (int64, error) {
ct := ©From{
conn: c,
tableName: tableName,
columnNames: columnNames,
rowSrc: rowSrc,
readerErrChan: make(chan error),
mode: c.config.DefaultQueryExecMode,
}
return ct.run(ctx)
}