Question about conccurrency control and Insert

Поиск
Список
Период
Сортировка
От Stéphane Cazeaux
Тема Question about conccurrency control and Insert
Дата
Msg-id 3F5ED3F7.2080900@netcentrex.net
обсуждение исходный текст
Ответы Re: Question about conccurrency control and Insert  (Richard Huxton <dev@archonet.com>)
Re: Question about conccurrency control and Insert  (Stephan Szabo <sszabo@megazone.bigpanda.com>)
Список pgsql-general
Дерево обсуждения
Hi

I currently use pgsql 7.2.4 (but the following has also been seen on pgsql 7.3.3) with a transaction level set to "read committed".
It do a lot of little tests to understand how concurrency control works. Let see this scenario:

We have a table named "test_count" and a field named "count"
The table contains 1 entry with count=1

Client 1:
    BEGIN;
    SELECT count FROM test_count FOR UPDATE; --> returns the only entry "1"
    ...

Client 2 :
    BEGIN;
    SELECT count FROM test_count FOR UPDATE; --> this query is blocked, ok
    ...

We continue :
 
Client 1:
    INSERT INTO test_count VALUES (2);
    COMMIT;

Client 2: (after commit of client 1)
    [The select that was blocked is now free. But the result is the first row containing "1". I'm surprised by this result]
    SELECT count FROM test_count;  --> now returns the two rows, on containing "1", the other containing "2"
    COMMIT;

So my question is : why the SELECT...FOR UPDATE of client 2, when unblocked, returns only one row, and a following SELECT in the same transaction returns two rows ? Is there a mechanisme I don't understand ?


Thanks for your response.
   
   

В списке pgsql-general по дате отправления:

Предыдущее
От: Tom Lane
Дата:
Сообщение: Re: Find overlapping time intervals, how?
Следующее
От: "Bruno BAGUETTE"
Дата:
Сообщение: RE : How to convert a UnixTimestamp to a PostgreSQL date without using ::abstime ?