HeapTupleHeaderData Layout description
<div style="font-family:Arial, Helvetica, sans-serif; font-size:12px;">Hi,<br>
<br>
In https://www.postgresql.org/docs/13/storage-page-layout.html<br>
we can find on § 68.6.1. Table Row Layout<br>
the Table 68.4. HeapTupleHeaderData Layout<br>
<br>
When additionning Length of differents Fields taht are present we obtain atotal of<br>
4+4+4+4+6+2+2+1 = 27<br>
<br>
What doesn't seem to fit with previous indication "There is a fixed-size header (occupying 23 bytes on most machines)"<br>
<br>
Researching in old doc shows that in 8.2 the documentation<br>
https://www.postgresql.org/docs/8.2/storage-page-layout.html<br>
indicate 27 bytes (with the same detailled Table)<br>
<br>
Number 23 appears from Postgres 8.3<br>
<br>
In some thread like https://stackoverflow.com/questions/2966524/calculating-and-saving-space-in-postgresql/7431468<br>
It's possible to read that<br>
Overhead per tuple (row)<br>
... And at least 24 bytes (23 + padding) for the tuple header ...<br>
<br>
As there is several thread about this point I suppose there is an evolution between 8.2 and 8.3 with a missdocumentation for an evolution of detailled Table "HeapTupleHeaderData Layout"<br>
<br>
IS it right ? And in this case what is the real detailled Table ?<br>
<br>
Thanks for any explanations<br>
Regards</div>
benj.dev@laposte.net writes:
In https://www.postgresql.org/docs/13/storage-page-layout.html<br>
we can find on § 68.6.1. Table Row Layout<br>
the Table 68.4. HeapTupleHeaderData Layout<br>
<br>
When additionning Length of differents Fields taht are present we obtain atotal of<br>
4+4+4+4+6+2+2+1 = 27<br>
I think you missed the point that t_cid overlays with t_xvac.
regards, tom lane
Le 22/01/2021 à 17:49, Tom Lane a écrit :
benj.dev@laposte.net writes:
In https://www.postgresql.org/docs/13/storage-page-layout.html<br>
we can find on § 68.6.1. Table Row Layout<br>
the Table 68.4. HeapTupleHeaderData Layout<br>
<br>
When additionning Length of differents Fields taht are present we obtain atotal of<br>
4+4+4+4+6+2+2+1 = 27<br>I think you missed the point that t_cid overlays with t_xvac.
regards, tom lane
Hi,
Thank you for this point. It's more clear.
So in fact the error (with number 27) was in the documentation for
version before postgres 8.3.
In version >=8.3, the overlays precision could be added into sentence
"XID for VACUUM operation moving a row version" => "XID for VACUUM
operation moving a row version (overlays with t_cid)"
or
Maybe show t_cid and t_xvac in the same cell into the array could be
more clear
Example :
|===========|===============|=========|================================|
|Field |Type |Length |Description |
|===========|===============|=========|================================|
|t_xmin |TransactionId |4 octets |insert XID stamp |
|-----------|---------------|---------|--------------------------------|
|t_xmax |TransactionId |4 octets |delete XID stamp |
|-----------|---------------|---------|--------------------------------|
|t_cid |CommandId | |insert and/or delete CID stamp |
|t_xvac |TransactionId | |XID for VACUUM operation moving |
| | | |a row version |
| | |4 octets |(t_cid and t_xvax are overlayed)|
|-----------|---------------|---------|--------------------------------|
|t_ctid |ItemPointerData|6 octets |current TID of this or newer row|
| | | |version |
|-----------|---------------|---------|--------------------------------|
|t_infomask2|uint16 |2 octets |number of attributes, plus |
| | | |various flag bits |
|-----------|---------------|---------|--------------------------------|
|t_infomask |uint16 |2 octets |various flag bits |
|-----------|---------------|---------|--------------------------------|
|t_hoff |uint8 |1 octet |offset to user data |
|===========|===============|=========|================================|
Thanks,
Regards
benj.dev@laposte.net writes:
Le 22/01/2021 à 17:49, Tom Lane a écrit :
I think you missed the point that t_cid overlays with t_xvac.
So in fact the error (with number 27) was in the documentation for
version before postgres 8.3.
No, the pre-8.3 docs are also correct, for their time.
regards, tom lane