Re: Add new function to convert 32-bit XID to 64-bit - Mailing list pgsql-hackers

From Ashutosh Bapat
Subject Re: Add new function to convert 32-bit XID to 64-bit
Date
Msg-id CAExHW5tQ+=GA8EJojqXR_0Boh-x0iqNptprY=PHkUkmCxqcdCA@mail.gmail.com
Whole thread Raw
In response to Add new function to convert 32-bit XID to 64-bit  (Fujii Masao <masao.fujii@oss.nttdata.com>)
List pgsql-hackers
On Thu, Dec 23, 2021 at 8:33 PM Fujii Masao <masao.fujii@oss.nttdata.com> wrote:
>
> Hi,
>
> When periodically collecting and accumulating statistics or status information like pg_locks, pg_stat_activity,
pg_prepared_xacts,etc for future troubleshooting or some reasons, I'd like to store a transaction ID of such
informationas 64-bit version so that the information of specified transaction easily can be identified and picked up by
transactionID. Otherwise it's not easy to distinguish transactions with the same 32-bit XID but different epoch, only
by32-bit XID. 
>
> But since pg_locks or pg_stat_activity etc returns 32-bit XID, we could not store it as 64-bit version. To improve
thissituation, I'd like to propose to add new function that converts the given 32-bit XID (i.e., xid data type) to
64-bitone (xid8 data type). If we do this, for example we can easily get 64-bit XID from pg_stat_activity by "SELECT
convert_xid_to_xid8(backend_xid)FROM pg_stat_activity", if necessary. Thought? 

What will this function do?

From your earlier description it looks like you want this function to
add epoch to the xid when making it a 64bit value. Is that right?

>
> As another approach, we can change the data type of pg_stat_activity.backend_xid or pg_locks.transaction, etc from
xidto xid8. But this idea looks overkill to me, and it may break the existing applications accessing pg_locks etc. So
IMOit's better to just provide the convertion function. 

Obviously we will break backward compatibility for applications
upgrading to a newer PostgreSQL version. Downside is applications
using 64bit xid will need to change their applications.

If we want to change the datatype anyway, better to create a new type
LongTransactionId or something like to represent 64bit transaction id
and then change these functions to use that.

--
Best Wishes,
Ashutosh Bapat



pgsql-hackers by date:

Previous
From: Bharath Rupireddy
Date:
Subject: Re: correct the sizes of values and nulls arrays in pg_control_checkpoint
Next
From: "houzj.fnst@fujitsu.com"
Date:
Subject: RE: Delay the variable initialization in get_rel_sync_entry