You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Copy file name to clipboardExpand all lines: README.md
+41Lines changed: 41 additions & 0 deletions
Display the source diff
Display the rich diff
Original file line number
Diff line number
Diff line change
@@ -11,6 +11,8 @@ It does two things:
11
11
* convert a SQL Server schema to a PostgreSQL schema
12
12
* produce a Pentaho Data Integrator (Kettle) job to migrate
13
13
all the data from SQL Server to PostgreSQL. This second part is optionnal
14
+
* produce an incremental version of this job to migrate what has changed in the database from the previous run. This is created
15
+
when the migration job is also created.
14
16
15
17
16
18
Notes, warnings:
@@ -90,6 +92,34 @@ the size of the scale of the numeric
90
92
if no: they are created as not valid (enforced only for new rows)
91
93
if after: they are created as not valid, but the statements to validate them are put in the unsure file
92
94
95
+
-sort_size=100000: sort size to use for incremental jobs. Default is 10000, to try to be on the safe side (see below).
96
+
97
+
We don't sort in databases for two reasons: the sort order can be different between SQL Server
98
+
and PostgreSQL, and we don't want to stress the servers. But sorting a lot of data in Java can generate a Java Out of Heap Memory error.
99
+
If you get Out of Memory errors, raise the Java Heap memory (in the kitchen script) as much as you can. If you still have the problem, reduce
100
+
this sort size. You can also try reducing parallelism, having one on two sorts instead of 8 will of course consume less memory. The last problem is that
101
+
if the sort_size is small, kettle is going to generate a very large amount of temporary files, and then read them back sorted. So you may hit the
102
+
"too many open files" limit of your system (default 1024 on linux for instance). So you'll have to do some tuning here:
103
+
- First, use as much Java memory as you can: set the JAVAXMEM environment variable to 4096 (megabytes) or more if you can afford it. The more the better.
104
+
- If you still get Out Of Memory errors, put a smaller sort size, until you can do the sorts (decrease it tenfold each time for example). You'll obviously lose some performance
105
+
- If then you get the too many open files error, raise the maximum number of open files. In most Linux distributions, this is editing /etc/security/limits.conf and putting
106
+
@userName soft nofile 65535
107
+
@userName hard nofile 65535
108
+
(replace userName with your user name). Log in again, and verify with "ulimit -n" that you are now allowed to open 65535 files.
109
+
You may also have to raise the maximum number of open files on the system: echo the new value to /proc/sys/fs/file-max.
110
+
111
+
You'll need a lot of temporary space on disk to do these sorts...
112
+
113
+
You can also edit only the offending transformation with Spoon (Kettle's GUI), so that only this one is slowed down.
114
+
115
+
When Kettle crashed on one of these problems, the temporary files aren't removed. They are usually in /tmp (or in your temp directory in Windows), and start with out_. Don't forget to remove them.
116
+
117
+
-use_pk_if_possible=0/1/public.table1,myschema.table2: enable the generation of jobs doing sorts in the databases (order by in the select part of Kettle's table inputs).
118
+
119
+
1 will ask to try for all tables, or you can give a list of tables (if for example, you cannot make these tables work with a reasonable sort size). Anyway, sqlserver2sql will only accept
120
+
to do sorts in the database if the primary key can be guaranteed to be sorted the same way in PostgreSQL and SQL Server. That means that it only accepts if the key is made only of numeric
121
+
and date/timestamp types. If not, the standard incremental job will be generated.
@@ -127,6 +157,17 @@ You've generated everything. Let's do the import:
127
157
128
158
If you want to dig deeper into the kettle job, you can use kettle_report.pl to display the individual table's transfer performance. Then, if needed, you'll be able to modify the Kettle job to optimize it, using Spoon, Kettle's GUI
This one is highly experimental. I need your feedback ! :). You should only run an incremental job on an already loaded database.
165
+
166
+
It may fail for a variety of reasons, mainly out of memory errors. If you have other unique constraints beyond the primary key, the series of queries generated by sqlserver2pgsql may generate conflicting updates. So test it several times before the migration day, if you really want to try this method. The "normal" method is safer, but of course, you'll start from scratch, and have those long indexes builds at the end.
167
+
168
+
By the way, to be able to insert data into all tables, it deactivates triggers at the beginning and activates them back at the end of the job. So, if the job fails, those triggers won't be reactivated.
169
+
170
+
```
130
171
You can also use a configuration file if you like:
0 commit comments