Message-ID: <58987a44-878a-338b-31de-4543c90604f1@dunslane.net>
Date: Wed, 3 Jan 2024 09:59:37 -0500
MIME-Version: 1.0
User-Agent: Mozilla/5.0 (X11; Linux x86_64; rv:102.0) Gecko/20100101
 Thunderbird/102.15.1
Content-Language: en-US
To: Robert Haas <robertmhaas@gmail.com>
Cc: PostgreSQL Hackers <pgsql-hackers@lists.postgresql.org>
References: <7b0a51d6-0d9d-7366-3a1a-f74397a02f55@dunslane.net>
 <CA+TgmoZKqZ+xXN550e2DXM-F4ELCE+Ee=Ogy4pZZMhu9SBRmzQ@mail.gmail.com>
 <253cef9e-1360-a1a6-5091-ad4368b1b8cc@dunslane.net>
 <CA+TgmoYLi7RjjbWvBHa_9+2rVTuO4C1xPHJqzCdNUnHxR3-NdA@mail.gmail.com>
From: Andrew Dunstan <andrew@dunslane.net>
Subject: Re: WIP Incremental JSON Parser
In-Reply-To: <CA+TgmoYLi7RjjbWvBHa_9+2rVTuO4C1xPHJqzCdNUnHxR3-NdA@mail.gmail.com>
Content-Type: text/plain; charset=UTF-8; format=flowed
Content-Transfer-Encoding: 8bit
Archived-At: <https://www.postgresql.org/message-id/58987a44-878a-338b-31de-4543c90604f1%40dunslane.net>
Precedence: bulk


On 2024-01-03 We 08:45, Robert Haas wrote:
> On Wed, Jan 3, 2024 at 6:57 AM Andrew Dunstan <andrew@dunslane.net> wrote:
>> Yeah. One idea I had yesterday was to stash the field names, which in
>> large JSON docs tent to be pretty repetitive, in a hash table instead of
>> pstrduping each instance. The name would be valid until the end of the
>> parse, and would only need to be duplicated by the callback function if
>> it were needed beyond that. That's not the case currently with the
>> parse_manifest code. I'll work on using a hash table.
> IMHO, this is not a good direction. Anybody who is parsing JSON
> probably wants to discard the duplicated labels and convert other
> heavily duplicated strings to enum values or something. (e.g. if every
> record has {"color":"red"} or {"color":"green"}). So the hash table
> lookups will cost but won't really save anything more than just
> freeing the memory not needed, but will probably be more expensive.


I don't quite follow.

Say we have a document with an array 1m objects, each with a field 
called "color". As it stands we'll allocate space for that field name 1m 
times. Using a hash table we'd allocated space for it once. And 
allocating the memory isn't free, although it might be cheaper than 
doing hash lookups.

I guess we can benchmark it and see what the performance impact of using 
a hash table might be.

Another possibility would be simply to have the callback free the field 
name after use. for the parse_manifest code that could be a one-line 
addition to the code at the bottom of json_object_manifest_field_start().


>> The parse_manifest code does seem to pfree the scalar values it no
>> longer needs fairly well, so maybe we don't need to to anything there.
> Hmm. This makes me wonder if you've measured how much actual leakage there is?


No I haven't. I have simply theorized about how much memory we might 
consume if nothing were done by the callers to free the memory.


cheers


andrew

--
Andrew Dunstan
EDB: https://www.enterprisedb.com