Обсуждение: XLOG's implementation details
Hi all, I am new to Postgres and am interested in XLOG's implementation details. I browsed around the source code, and saw that the particular function that appends XLOG entries is called from multiple places (heap, indexes, transaction subsystem, and so forth). Seeing this, I became particularly interested in the following: How strong is the coupling between XLOG entries and database pages? That is, assume that I am inserting one row to a database table. Is there anyway for me to construct the related XLOG entry before calling the heap insert function? Or, are XLOG files conceptually binary diffs for database pages, moving forward in time? Another question that I had relates to XLOG files and tables. If I have two databases and one is set up as a slave to the other, is it possible for the slave to have fewer tables than the master? If it is possible, is that because tables don't share pages? Again I'm new to Postgres, and am curious about how stuff works underneath the covers. I figured asking this list would be the simplest way to figure things out. Thanks, Ozgun.
On Nov 13, 2009, at 3:46 AM, Ozgun Erdogan wrote: > How strong is the coupling between XLOG entries and database pages? > That is, assume that I am inserting one row to a database table. Is > there anyway for me to construct the related XLOG entry before calling > the heap insert function? Or, are XLOG files conceptually binary diffs > for database pages, moving forward in time? The latter. The converse of your question is much more common, "How can I reconstruct the statements that changed my database from the xlog contents", with the same answer. > Another question that I had relates to XLOG files and tables. If I > have two databases and one is set up as a slave to the other, is it > possible for the slave to have fewer tables than the master? If it is > possible, is that because tables don't share pages? That depends on how you implement replication. Since you ask in the context of xlogs I'll assume you mean a warm standby slave via WAL shipping and there the answer is no. Well, you might be able to work something out by using record-based log shipping but even if that could work it would take a lot of, well, work. > Again I'm new to Postgres, and am curious about how stuff works > underneath the covers. I figured asking this list would be the > simplest way to figure things out. Yep :) Erik Jones, Database Administrator Engine Yard Support, Scalability, Reliability 866.518.9273 x 260 Location: US/Pacific IRC: mage2k
On Nov 13, 2009, at 3:46 AM, Ozgun Erdogan wrote: > How strong is the coupling between XLOG entries and database pages? > That is, assume that I am inserting one row to a database table. Is > there anyway for me to construct the related XLOG entry before calling > the heap insert function? Or, are XLOG files conceptually binary diffs > for database pages, moving forward in time? The latter. The converse of your question is much more common, "How can I reconstruct the statements that changed my database from the xlog contents", with the same answer. > Another question that I had relates to XLOG files and tables. If I > have two databases and one is set up as a slave to the other, is it > possible for the slave to have fewer tables than the master? If it is > possible, is that because tables don't share pages? That depends on how you implement replication. Since you ask in the context of xlogs I'll assume you mean a warm standby slave via WAL shipping and there the answer is no. Well, you might be able to work something out by using record-based log shipping but even if that could work it would take a lot of, well, work. > Again I'm new to Postgres, and am curious about how stuff works > underneath the covers. I figured asking this list would be the > simplest way to figure things out. Yep :) Erik Jones, Database Administrator Engine Yard Support, Scalability, Reliability 866.518.9273 x 260 Location: US/Pacific IRC: mage2k