1
|
# Database import/export
|
2
|
|
3
|
import exc
|
4
|
import dicts
|
5
|
import sql
|
6
|
import sql_gen
|
7
|
import strings
|
8
|
import util
|
9
|
|
10
|
##### Data cleanup
|
11
|
|
12
|
def cleanup_table(db, table, cols):
|
13
|
table = sql_gen.as_Table(table)
|
14
|
cols = map(sql_gen.as_Col, cols)
|
15
|
|
16
|
expr = ('nullif(nullif(trim(both from %s), '+db.esc_value('')+'), '
|
17
|
+db.esc_value(r'\N')+')')
|
18
|
changes = [(v, sql_gen.CustomCode(expr % v.to_str(db)))
|
19
|
for v in cols]
|
20
|
|
21
|
sql.update(db, table, changes, in_place=True)
|
22
|
|
23
|
##### Error tracking
|
24
|
|
25
|
def track_data_error(db, errors_table, cols, value, error_code, error):
|
26
|
'''
|
27
|
@param errors_table If None, does nothing.
|
28
|
'''
|
29
|
if errors_table == None or cols == (): return
|
30
|
|
31
|
for col in cols:
|
32
|
try:
|
33
|
sql.insert(db, errors_table, dict(column=col.name, value=value,
|
34
|
error_code=error_code, error=error), recover=True,
|
35
|
cacheable=True, log_level=4)
|
36
|
except sql.DuplicateKeyException: pass
|
37
|
|
38
|
def cast(db, type_, col, errors_table=None):
|
39
|
'''Casts an (unrenamed) column or value.
|
40
|
If errors_table set and col has srcs, saves errors in errors_table (using
|
41
|
col's srcs attr as the source columns) and converts errors to warnings.
|
42
|
@param col str|sql_gen.Col|sql_gen.Literal
|
43
|
@param errors_table None|sql_gen.Table|str
|
44
|
'''
|
45
|
col = sql_gen.as_Col(col)
|
46
|
save_errors = (errors_table != None and isinstance(col, sql_gen.Col)
|
47
|
and col.srcs != ())
|
48
|
if not save_errors: return sql_gen.Cast(type_, col) # can't save errors
|
49
|
|
50
|
assert not isinstance(col, sql_gen.NamedCol)
|
51
|
|
52
|
errors_table = sql_gen.as_Table(errors_table)
|
53
|
srcs = map(sql_gen.to_name_only_col, col.srcs)
|
54
|
function_name = str(sql_gen.FunctionCall(type_, *srcs))
|
55
|
function = db.TempFunction(function_name)
|
56
|
|
57
|
while True:
|
58
|
# Create function definition
|
59
|
errors_table_cols = map(sql_gen.Col,
|
60
|
['column', 'value', 'error_code', 'error'])
|
61
|
query = '''\
|
62
|
CREATE FUNCTION '''+function.to_str(db)+'''(value text)
|
63
|
RETURNS '''+type_+'''
|
64
|
LANGUAGE plpgsql
|
65
|
STRICT
|
66
|
AS $$
|
67
|
BEGIN
|
68
|
/* The explicit cast to the return type is needed to make the cast happen
|
69
|
inside the try block. (Implicit casts to the return type happen at the end
|
70
|
of the function, outside any block.) */
|
71
|
RETURN value::'''+type_+''';
|
72
|
EXCEPTION
|
73
|
WHEN data_exception THEN
|
74
|
-- Save error in errors table.
|
75
|
DECLARE
|
76
|
error_code text := SQLSTATE;
|
77
|
error text := SQLERRM;
|
78
|
"column" text;
|
79
|
BEGIN
|
80
|
-- Insert the value and error for *each* source column.
|
81
|
FOR "column" IN
|
82
|
'''+sql.mk_select(db, sql_gen.NamedValues('c', None, [[c.name] for c in srcs]),
|
83
|
order_by=None, start=0)+'''
|
84
|
LOOP
|
85
|
BEGIN
|
86
|
'''+sql.mk_insert_select(db, errors_table, errors_table_cols,
|
87
|
sql_gen.Values(errors_table_cols).to_str(db))+''';
|
88
|
EXCEPTION
|
89
|
WHEN unique_violation THEN NULL; -- continue to next row
|
90
|
END;
|
91
|
END LOOP;
|
92
|
END;
|
93
|
|
94
|
RAISE WARNING '%', SQLERRM;
|
95
|
RETURN NULL;
|
96
|
END;
|
97
|
$$;
|
98
|
'''
|
99
|
|
100
|
# Create function
|
101
|
try:
|
102
|
sql.run_query(db, query, recover=True, cacheable=True,
|
103
|
log_ignore_excs=(sql.DuplicateException,))
|
104
|
break # successful
|
105
|
except sql.DuplicateException:
|
106
|
function.name = sql.next_version(function.name)
|
107
|
# try again with next version of name
|
108
|
|
109
|
return sql_gen.FunctionCall(function, col)
|
110
|
|
111
|
def cast_temp_col(db, type_, col, errors_table=None):
|
112
|
'''Like cast(), but creates a new column with the cast values if the input
|
113
|
is a column.
|
114
|
@return The new column or cast value
|
115
|
'''
|
116
|
def cast_(col): return cast(db, type_, col, errors_table)
|
117
|
|
118
|
try: col = sql_gen.underlying_col(col)
|
119
|
except sql_gen.NoUnderlyingTableException: return sql_gen.wrap(cast_, col)
|
120
|
|
121
|
table = col.table
|
122
|
new_col = sql_gen.Col(sql_gen.concat(col.name, '::'+type_), table, col.srcs)
|
123
|
expr = cast_(col)
|
124
|
|
125
|
# Add column
|
126
|
new_typed_col = sql_gen.TypedCol(new_col.name, type_)
|
127
|
sql.add_col(db, table, new_typed_col, comment='src: '+repr(col))
|
128
|
new_col.name = new_typed_col.name # propagate any renaming
|
129
|
|
130
|
sql.update(db, table, [(new_col, expr)], in_place=True, cacheable=True)
|
131
|
sql.add_index(db, new_col)
|
132
|
|
133
|
return new_col
|
134
|
|
135
|
def errors_table(db, table, if_exists=True):
|
136
|
'''
|
137
|
@param if_exists If set, returns None if the errors table doesn't exist
|
138
|
@return None|sql_gen.Table
|
139
|
'''
|
140
|
table = sql_gen.as_Table(table)
|
141
|
if table.srcs != (): table = table.srcs[0]
|
142
|
|
143
|
errors_table = sql_gen.suffixed_table(table, '.errors')
|
144
|
if if_exists and not sql.table_exists(db, errors_table): return None
|
145
|
return errors_table
|
146
|
|
147
|
##### Import
|
148
|
|
149
|
def put(db, table, row, pkey_=None, row_ct_ref=None):
|
150
|
'''Recovers from errors.
|
151
|
Only works under PostgreSQL (uses INSERT RETURNING).
|
152
|
'''
|
153
|
row = sql_gen.ColDict(db, table, row)
|
154
|
if pkey_ == None: pkey_ = sql.pkey(db, table, recover=True)
|
155
|
|
156
|
try:
|
157
|
cur = sql.insert(db, table, row, pkey_, recover=True)
|
158
|
if row_ct_ref != None and cur.rowcount >= 0:
|
159
|
row_ct_ref[0] += cur.rowcount
|
160
|
return sql.value(cur)
|
161
|
except sql.DuplicateKeyException, e:
|
162
|
row = sql_gen.ColDict(db, table,
|
163
|
util.dict_subset_right_join(row, e.cols))
|
164
|
return sql.value(sql.select(db, table, [pkey_], row, recover=True))
|
165
|
|
166
|
def get(db, table, row, pkey, row_ct_ref=None, create=False):
|
167
|
'''Recovers from errors'''
|
168
|
try:
|
169
|
return sql.value(sql.select(db, table, [pkey], row, limit=1,
|
170
|
recover=True))
|
171
|
except StopIteration:
|
172
|
if not create: raise
|
173
|
return put(db, table, row, pkey, row_ct_ref) # insert new row
|
174
|
|
175
|
def is_func_result(col):
|
176
|
return col.table.name.find('(') >= 0 and col.name == 'result'
|
177
|
|
178
|
def into_table_name(out_table, in_tables0, mapping, is_func):
|
179
|
def in_col_str(in_col):
|
180
|
in_col = sql_gen.remove_col_rename(in_col)
|
181
|
if isinstance(in_col, sql_gen.Col):
|
182
|
table = in_col.table
|
183
|
if table == in_tables0:
|
184
|
in_col = sql_gen.to_name_only_col(in_col)
|
185
|
elif is_func_result(in_col): in_col = table # omit col name
|
186
|
return str(in_col)
|
187
|
|
188
|
str_ = str(out_table)
|
189
|
if is_func:
|
190
|
str_ += '('
|
191
|
|
192
|
try: value_in_col = mapping['value']
|
193
|
except KeyError:
|
194
|
str_ += ', '.join((str(k)+'='+in_col_str(v)
|
195
|
for k, v in mapping.iteritems()))
|
196
|
else: str_ += in_col_str(value_in_col)
|
197
|
|
198
|
str_ += ')'
|
199
|
else:
|
200
|
out_col = 'rank'
|
201
|
try: in_col = mapping[out_col]
|
202
|
except KeyError: str_ += '_pkeys'
|
203
|
else: # has a rank column, so hierarchical
|
204
|
str_ += '['+str(out_col)+'='+in_col_str(in_col)+']'
|
205
|
return str_
|
206
|
|
207
|
def put_table(db, out_table, in_tables, mapping, row_ct_ref=None, into=None,
|
208
|
default=None, is_func=False, on_error=exc.raise_):
|
209
|
'''Recovers from errors.
|
210
|
Only works under PostgreSQL (uses INSERT RETURNING).
|
211
|
IMPORTANT: Must be run at the *beginning* of a transaction.
|
212
|
@param in_tables The main input table to select from, followed by a list of
|
213
|
tables to join with it using the main input table's pkey
|
214
|
@param mapping dict(out_table_col=in_table_col, ...)
|
215
|
* out_table_col: str (*not* sql_gen.Col)
|
216
|
* in_table_col: sql_gen.Col|literal-value
|
217
|
@param into The table to contain the output and input pkeys.
|
218
|
Defaults to `out_table.name+'_pkeys'`.
|
219
|
@param default The *output* column to use as the pkey for missing rows.
|
220
|
If this output column does not exist in the mapping, uses None.
|
221
|
@param is_func Whether out_table is the name of a SQL function, not a table
|
222
|
@return sql_gen.Col Where the output pkeys are made available
|
223
|
'''
|
224
|
out_table = sql_gen.as_Table(out_table)
|
225
|
|
226
|
def log_debug(msg): db.log_debug(msg, level=1.5)
|
227
|
def col_ustr(str_):
|
228
|
return strings.repr_no_u(sql_gen.remove_col_rename(str_))
|
229
|
|
230
|
log_debug('********** New iteration **********')
|
231
|
log_debug('Inserting these input columns into '+strings.as_tt(
|
232
|
out_table.to_str(db))+':\n'+strings.as_table(mapping, ustr=col_ustr))
|
233
|
|
234
|
is_function = sql.function_exists(db, out_table)
|
235
|
|
236
|
if is_function: out_pkey = 'result'
|
237
|
else: out_pkey = sql.pkey(db, out_table, recover=True)
|
238
|
out_pkey_col = sql_gen.as_Col(out_pkey, out_table)
|
239
|
|
240
|
if mapping == {}: # need at least one column for INSERT SELECT
|
241
|
mapping = {out_pkey: None} # ColDict will replace with default value
|
242
|
|
243
|
# Create input joins from list of input tables
|
244
|
in_tables_ = in_tables[:] # don't modify input!
|
245
|
in_tables0 = in_tables_.pop(0) # first table is separate
|
246
|
errors_table_ = errors_table(db, in_tables0)
|
247
|
in_pkey = sql.pkey(db, in_tables0, recover=True)
|
248
|
in_pkey_col = sql_gen.as_Col(in_pkey, in_tables0)
|
249
|
input_joins = [in_tables0]+[sql_gen.Join(v,
|
250
|
{in_pkey: sql_gen.join_same_not_null}) for v in in_tables_]
|
251
|
|
252
|
if into == None:
|
253
|
into = into_table_name(out_table, in_tables0, mapping, is_func)
|
254
|
into = sql_gen.as_Table(into)
|
255
|
|
256
|
# Set column sources
|
257
|
in_cols = filter(sql_gen.is_table_col, mapping.values())
|
258
|
for col in in_cols:
|
259
|
if col.table == in_tables0: col.set_srcs(sql_gen.src_self)
|
260
|
|
261
|
log_debug('Joining together input tables into temp table')
|
262
|
# Place in new table for speed and so don't modify input if values edited
|
263
|
in_table = sql_gen.Table('in')
|
264
|
mapping = dicts.join(mapping, sql.flatten(db, in_table, input_joins,
|
265
|
in_cols, preserve=[in_pkey_col], start=0))
|
266
|
input_joins = [in_table]
|
267
|
db.log_debug('Temp table: '+strings.as_tt(in_table.to_str(db)), level=2)
|
268
|
|
269
|
mapping = sql_gen.ColDict(db, out_table, mapping)
|
270
|
# after applying dicts.join() because that returns a plain dict
|
271
|
|
272
|
# Resolve default value column
|
273
|
if default != None:
|
274
|
try: default = mapping[default]
|
275
|
except KeyError:
|
276
|
db.log_debug('Default value column '
|
277
|
+strings.as_tt(strings.repr_no_u(default))
|
278
|
+' does not exist in mapping, falling back to None', level=2.1)
|
279
|
default = None
|
280
|
|
281
|
pkeys_names = [in_pkey, out_pkey]
|
282
|
pkeys_cols = [in_pkey_col, out_pkey_col]
|
283
|
|
284
|
pkeys_table_exists_ref = [False]
|
285
|
def insert_into_pkeys(joins, cols, distinct=False):
|
286
|
kw_args = {}
|
287
|
if distinct: kw_args.update(dict(distinct_on=[in_pkey_col]))
|
288
|
query = sql.mk_select(db, joins, cols, order_by=None, start=0,
|
289
|
**kw_args)
|
290
|
|
291
|
if pkeys_table_exists_ref[0]:
|
292
|
sql.insert_select(db, into, pkeys_names, query)
|
293
|
else:
|
294
|
sql.run_query_into(db, query, into=into)
|
295
|
pkeys_table_exists_ref[0] = True
|
296
|
|
297
|
limit_ref = [None]
|
298
|
insert_in_table = in_table
|
299
|
conds = set()
|
300
|
distinct_on = sql_gen.ColDict(db, out_table)
|
301
|
def mk_main_select(joins, cols):
|
302
|
conds_ = [(sql_gen.with_table(k, insert_in_table), v) for k, v in conds]
|
303
|
return sql.mk_select(db, joins, cols, conds_, limit=limit_ref[0],
|
304
|
start=0)
|
305
|
|
306
|
exc_strs = set()
|
307
|
def log_exc(e):
|
308
|
e_str = exc.str_(e, first_line_only=True)
|
309
|
log_debug('Caught exception: '+e_str)
|
310
|
assert e_str not in exc_strs # avoid infinite loops
|
311
|
exc_strs.add(e_str)
|
312
|
|
313
|
def remove_all_rows():
|
314
|
log_debug('Ignoring all rows')
|
315
|
limit_ref[0] = 0 # just create an empty pkeys table
|
316
|
|
317
|
def ignore(in_col, value, e):
|
318
|
track_data_error(db, errors_table_, in_col.srcs, value,
|
319
|
e.cause.pgcode, e.cause.pgerror)
|
320
|
log_debug('Ignoring rows with '+strings.as_tt(repr(in_col))+' = '
|
321
|
+strings.as_tt(repr(value)))
|
322
|
|
323
|
def remove_rows(in_col, value, e):
|
324
|
ignore(in_col, value, e)
|
325
|
cond = (in_col, sql_gen.CompareCond(value, '!='))
|
326
|
assert cond not in conds # avoid infinite loops
|
327
|
conds.add(cond)
|
328
|
|
329
|
def invalid2null(in_col, value, e):
|
330
|
ignore(in_col, value, e)
|
331
|
sql.update(db, in_table, [(in_col, None)],
|
332
|
sql_gen.ColValueCond(in_col, value))
|
333
|
|
334
|
def insert_pkeys_table(which):
|
335
|
return sql_gen.Table(sql_gen.concat(in_table.name,
|
336
|
'_insert_'+which+'_pkeys'))
|
337
|
insert_out_pkeys = insert_pkeys_table('out')
|
338
|
insert_in_pkeys = insert_pkeys_table('in')
|
339
|
|
340
|
# Do inserts and selects
|
341
|
join_cols = sql_gen.ColDict(db, out_table)
|
342
|
while True:
|
343
|
if limit_ref[0] == 0: # special case
|
344
|
log_debug('Creating an empty pkeys table')
|
345
|
cur = sql.run_query_into(db, sql.mk_select(db, out_table,
|
346
|
[out_pkey], limit=limit_ref[0]), into=insert_out_pkeys)
|
347
|
break # don't do main case
|
348
|
|
349
|
has_joins = join_cols != {}
|
350
|
|
351
|
log_debug('Trying to insert new rows')
|
352
|
|
353
|
# Prepare to insert new rows
|
354
|
insert_args = dict(recover=True, cacheable=False)
|
355
|
if has_joins:
|
356
|
insert_args.update(dict(ignore=True))
|
357
|
else:
|
358
|
insert_args.update(dict(returning=out_pkey, into=insert_out_pkeys))
|
359
|
main_select = mk_main_select([insert_in_table],
|
360
|
[sql_gen.with_table(c, insert_in_table) for c in mapping.values()])
|
361
|
|
362
|
def main_insert():
|
363
|
if is_function:
|
364
|
log_debug('Calling function on input rows')
|
365
|
args = dict(((k.name, v) for k, v in mapping.iteritems()))
|
366
|
func_call = sql_gen.NamedCol(out_pkey,
|
367
|
sql_gen.FunctionCall(out_table, **args))
|
368
|
insert_into_pkeys(input_joins, [in_pkey_col, func_call])
|
369
|
return None
|
370
|
else:
|
371
|
return sql.insert_select(db, out_table, mapping.keys(),
|
372
|
main_select, **insert_args)
|
373
|
|
374
|
try:
|
375
|
cur = sql.with_savepoint(db, main_insert)
|
376
|
break # insert successful
|
377
|
except sql.MissingCastException, e:
|
378
|
log_exc(e)
|
379
|
|
380
|
out_col = e.col
|
381
|
type_ = e.type
|
382
|
|
383
|
log_debug('Casting '+strings.as_tt(out_col)+' input to '
|
384
|
+strings.as_tt(type_))
|
385
|
mapping[out_col] = cast_temp_col(db, type_, mapping[out_col],
|
386
|
errors_table_)
|
387
|
except sql.DuplicateKeyException, e:
|
388
|
log_exc(e)
|
389
|
|
390
|
old_join_cols = join_cols.copy()
|
391
|
distinct_on.update(util.dict_subset(mapping, e.cols))
|
392
|
join_cols.update(util.dict_subset_right_join(mapping, e.cols))
|
393
|
log_debug('Ignoring existing rows, comparing on these columns:\n'
|
394
|
+strings.as_inline_table(join_cols, ustr=col_ustr))
|
395
|
assert join_cols != old_join_cols # avoid infinite loops
|
396
|
|
397
|
# Uniquify input table to avoid internal duplicate keys
|
398
|
insert_in_table = sql.distinct_table(db, insert_in_table,
|
399
|
filter(sql_gen.is_table_col, distinct_on.values()))
|
400
|
except sql.NullValueException, e:
|
401
|
log_exc(e)
|
402
|
|
403
|
out_col, = e.cols
|
404
|
try: in_col = mapping[out_col]
|
405
|
except KeyError:
|
406
|
log_debug('Missing mapping for NOT NULL column '+out_col)
|
407
|
remove_all_rows()
|
408
|
else: remove_rows(in_col, None, e)
|
409
|
except sql.DatabaseErrors, e:
|
410
|
log_exc(e)
|
411
|
|
412
|
log_debug('No handler for exception')
|
413
|
on_error(e)
|
414
|
remove_all_rows()
|
415
|
# after exception handled, rerun loop with additional constraints
|
416
|
|
417
|
if cur != None and row_ct_ref != None and cur.rowcount >= 0:
|
418
|
row_ct_ref[0] += cur.rowcount
|
419
|
|
420
|
if is_function: pass # pkeys table already created
|
421
|
elif has_joins:
|
422
|
select_joins = input_joins+[sql_gen.Join(out_table, join_cols)]
|
423
|
log_debug('Getting output table pkeys of existing/inserted rows')
|
424
|
insert_into_pkeys(select_joins, pkeys_cols, distinct=True)
|
425
|
else:
|
426
|
sql.add_row_num(db, insert_out_pkeys) # for joining with input pkeys
|
427
|
|
428
|
log_debug('Getting input table pkeys of inserted rows')
|
429
|
sql.run_query_into(db, mk_main_select(input_joins, [in_pkey]),
|
430
|
into=insert_in_pkeys)
|
431
|
sql.add_row_num(db, insert_in_pkeys) # for joining with output pkeys
|
432
|
|
433
|
assert sql.table_row_count(db, insert_out_pkeys) == sql.table_row_count(
|
434
|
db, insert_in_pkeys)
|
435
|
|
436
|
log_debug('Combining output and input pkeys in inserted order')
|
437
|
pkey_joins = [insert_in_pkeys, sql_gen.Join(insert_out_pkeys,
|
438
|
{sql.row_num_col: sql_gen.join_same_not_null})]
|
439
|
insert_into_pkeys(pkey_joins, pkeys_names)
|
440
|
|
441
|
sql.empty_temp(db, [insert_out_pkeys, insert_in_pkeys])
|
442
|
|
443
|
db.log_debug('Adding pkey on pkeys table to enable fast joins', level=2.5)
|
444
|
sql.add_pkey(db, into)
|
445
|
|
446
|
log_debug('Setting pkeys of missing rows to '+strings.as_tt(repr(default)))
|
447
|
missing_rows_joins = input_joins+[sql_gen.Join(into,
|
448
|
{in_pkey: sql_gen.join_same_not_null}, sql_gen.filter_out)]
|
449
|
# must use join_same_not_null or query will take forever
|
450
|
insert_into_pkeys(missing_rows_joins,
|
451
|
[in_pkey_col, sql_gen.NamedCol(out_pkey, default)])
|
452
|
|
453
|
assert sql.table_row_count(db, into) == sql.table_row_count(db, in_table)
|
454
|
|
455
|
sql.empty_temp(db, set([in_table, insert_in_table]))
|
456
|
|
457
|
srcs = []
|
458
|
if is_func: srcs = sql_gen.cols_srcs(in_cols)
|
459
|
return sql_gen.Col(out_pkey, into, srcs)
|