sds transactions

I don't understand sds secondary isolation in transactions I'm making some test and I'm a bit confused, perhaps something is not well set: --Primary create table kk(k char(1));begin work; insert into kk values (2); --Secondary select * from kk --> 2 --- --dirty read?; I keep test without closing transaction: --- --Primary delete from kk where 1 = 1; --Secondary select * from kk --> 2 --- --? didnt get the delete transaction --- --Primary insert into kk values (1); --Secondary select * from kk --> 1; --Primary insert into kk values (3); insert into kk values (4); ---Secondary select * from kk --> 1; ---- --????? ---- If you commit transactions in sds secondary you see all ok, it just when transactions are created in primary node and secondary queries the tables affected by the transaction. SDS read only in my tests...

Original post: I don't understand sds secondary isolation in transactions I'm making some test and I'm a bit confused, perhaps something is not well set: <steps removed> If you commit transactions in sds secondary you see all ok, it just when transactions are created in primary node and secondary queries the tables affected by the transaction. SDS read only in my tests... Response: I followed your steps and got the same results on 12.10.xC7. So first off, if you use onstat -g ses on the SDS instance for the session doing the queries, you can see that, yes the isolation level being used is dirty read. However, the oddities you are picking up, appears to be whether or not the log record for the DML action (so your inserts/deletes) is getting applied on the SDS server or not yet. When doing the inserts and deletes, the log records are in the logical log buffer, and may not immediately be flushed to disk (which is 1 of the mechanisms used to alert the SDS server it has log records it needs to apply so that the pages on both instances appears the same). So there can be a delay at this time where the change has happened on the primary, but the log record isn't yet applied on the SDS server. That would appear to be what you are seeing. If you throw in a 2nd window on your primary and did something that forced log buffer flushes (like an easy thing to do would be to force checkpoints after each insert/delete via onmode -c), then I believe you would see the results you would be expecting in your selects, as if they were the default of dirty read isolation level. If I remember correctly off the top of my head, I think to be able to use any other isolation level then dirty read on the SDS server, it needs to be updatable. Jacques Renaut IBM Informix Advanced Support