[Tarantool-patches] [PATCH 4/4] sql: provide a user friendly frontend for accessing session settings

[Chris Sosnin]

> On 17 Mar 2020, at 23:12, Nikita Pettik <korablev at tarantool.org> wrote:
> 
> On 17 Mar 20:26, Chris Sosnin wrote:
>> 
>> 
>>> On 16 Mar 2020, at 20:02, Nikita Pettik <korablev at tarantool.org> wrote:
>>> 
>>> On 17 Feb 15:12, Chris Sosnin wrote:
>>>> Currently if a user wants to change session setting with sql, he has
>>>> to execute non-obvious query, thus, we introduce a more native way to
>>>> do this.
>>> 
>>> It is not about non-obvious queries, but it is all about documentation:
>>> the better feature is described, the clearer its usage turns out to be
>>> for user.
>> 
>> Should I change the commit message? I though this patchset is about simplifying
>> the way session settings are updated?
> 
> I would say:
> 
> Currently if a user wants to change session settings via SQL, one has
> to execute UPDATE query like:
> [[UPDATE "_session_settings" SET "value" = true WHERE "name" = 'name']]
> However, direct access to system spaces isn't considered to be good practice.
> To avoid that and a bit simplify user's life, we introduce SQL shortcut command
> SET with following syntax: SET <setting_name> = <value>. <setting_name> is
> supposed to be name of setting (note that it is uppercased as any other
> non-quoted indentifiers in SQL) and <value> - value of corresponding setting to
> be set; <value> should be either literal or binding marker. Also it is worth
> noting that SET doesn't provide any implicit casts, so <value> must be of the
> type corresponding to the setting being updated. 
> 
> Example:
> ...
> 
> ^ It is up to you.

I took the beginning as long as the rest is described in the doc request (except for
implicit casts note, thanks for pointing that out, fixed):

Currently if a user wants to change session setting with SQL, one has
to execute UPDATE query like:
[[UPDATE "_session_settings" SET "value" = true WHERE "name" = 'name']]
However, direct access to system spaces isn't considered to be a good practice.
To avoid that and a bit simplify user's life, we introduce SQL shortcut command
SETTING SET.

> 
>>> 
>>>> Closes #4711
>>>> 
>>>> @TarantoolBot document
>>>> Title: API for accessing _session_settings space.
>>>> There are two ways of updating values of session settings:
>>>> via Lua and SQL.
>>>> 
>>>> SQL:
>>>> Instead of typing long UPDATE query one can use the SET statement:
>>>> `box.execute([[SET "<setting_name>" = <new_value>]])`.
>>>> Note, that this query is case sensitive so the name must be quoted.
>>>> 
>>>> Example:
>>>> ```
>>>> tarantool> box.execute([[set "sql_default_engine" = 'memtx']])
>>>> ---
>>>> - row_count: 1
>>>> ...
>>>> 
>>>> tarantool> box.execute([[set "sql_defer_foreign_keys" = true]])
>>> 
>>> Why didn't you consider show/get command to retrieve setting value?
>>> Otherwise, you simplify way to set session local options, but doesn't
>>> provide the same way to extract current values.
>> 
>> SELECT * FROM “_session_settings” is simple enough, isn’t it?
> 
> In this case you'll get list of all settings. To get specific one
> we should use this query:
> 
> SELECT value FROM "_session_settings" WHERE "name" = 'xxx'
> 
> I do not insist on implementing GET/SHOW SQL syntax now tho.
> 
>>>> diff --git a/src/box/sql/sqlInt.h b/src/box/sql/sqlInt.h
>>>> index d1fcf4761..3ffae5970 100644
>>>> --- a/src/box/sql/sqlInt.h
>>>> +++ b/src/box/sql/sqlInt.h
>>>> @@ -4510,4 +4510,15 @@ int
>>>> sql_fieldno_by_name(struct Parse *parse_context, struct Expr *field_name,
>>>> 		    uint32_t *fieldno);
>>>> 
>>>> diff --git a/src/box/sql/vdbe.c b/src/box/sql/vdbe.c
>>>> index 620d74e66..c81486fa6 100644
>>>> --- a/src/box/sql/vdbe.c
>>>> +++ b/src/box/sql/vdbe.c
>>>> @@ -55,6 +55,7 @@
>>>> #include "box/schema.h"
>>>> #include "box/space.h"
>>>> #include "box/sequence.h"
>>>> +#include "box/session_settings.h"
>>>> 
>>>> /*
>>>> * Invoke this macro on memory cells just prior to changing the
>>>> @@ -5248,6 +5249,55 @@ case OP_IncMaxid: {
>>>> 	break;
>>>> }
>>>> 
>>>> +/* Opcode: Set P1 P2 * * *
>>>> + *
>>>> + * Set the new value of the session setting. P1 is the id of the
>>>> + * setting in the session_settings array, P2 is the register
>>>> + * holding a value.
>>>> + */
>>>> +case OP_Set: {
>>> 
>>> Please, use more specific opcode name. Like OP_SettingSet.
>> 
>> Didn’t we accept ’set’ syntax?
> 
> SET is okay, but I don't mind SET SETTING as well. OP_Set is too general
> name - SET keyword is also part of UPDATE syntax, so it may confuse other
> developers. Except for this nit patch LGTM.


Here are the fixes:

-               sqlVdbeAddOp2(vdbe, OP_Set, index, target);
+              sqlVdbeAddOp2(vdbe, OP_SetSetting, index, target);

-///////////////////////////// The SET command ////////////////////////////////
-cmd ::= SET nm(X) EQ expr(Y).  {
+///////////////////////////// The SETTING SET command ////////////////////////
+cmd ::= SETTING SET nm(X) EQ expr(Y).  {

-/* Opcode: Set P1 P2 * * *
+/* Opcode: SetSetting P1 P2 * * *

-case OP_Set: {
+case OP_SetSetting: {

And of course I had to change the grammar and tests.

Changes are on the branch with the second version of Lua patch:
https://github.com/tarantool/tarantool/tree/ksosnin/gh-4712-session-settings-v2
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://lists.tarantool.org/pipermail/tarantool-patches/attachments/20200318/b0e456da/attachment.html>