Thread: About large objects asynchronous and non-blocking support
Hi. At the moment libpq doesn't seem to support asynchronous and non-blocking support for large objects, in the style of PQsendQuery/PQgetResult. This makes large objects hardly suited for single-threaded programs based on some variant of select(). I would like to know whether this is a deliberate decision or it is considered a bug, and, in case, whether it is scheduled to be fixed. Though I cannot guarantee anything, I may be interested into working out a patch, if no one is already doing the same (of course I understand that this patch wouldn't be for 9.3, which is already in its late release cycle). Do you think this may be of interest? Thanks, Giovanni. -- Giovanni Mascellani <mascellani@poisson.phc.unipi.it> Pisa, Italy Web: http://poisson.phc.unipi.it/~mascellani Jabber: g.mascellani@jabber.org / giovanni@elabor.homelinux.org
2013/6/5 Giovanni Mascellani <g.mascellani@gmail.com>
Hi.
At the moment libpq doesn't seem to support asynchronous and
non-blocking support for large objects, in the style of
PQsendQuery/PQgetResult. This makes large objects hardly suited for
single-threaded programs based on some variant of select().
"There are server-side functions callable from SQL that correspond to each of
the client-side functions". Hence, you can call these functions by using
asynchronous API.
// Dmitriy.
> Hi. > > At the moment libpq doesn't seem to support asynchronous and > non-blocking support for large objects, in the style of > PQsendQuery/PQgetResult. This makes large objects hardly suited for > single-threaded programs based on some variant of select(). > > I would like to know whether this is a deliberate decision or it is > considered a bug, and, in case, whether it is scheduled to be fixed. Certainly not bug, since the doc clearly stats that PQsendQuery can only be used as a substituation of PQexec. (see "Asynchronous Command Processing" section" for more details). The large object API is completely different from PQexec and its friends, so it cannot be used with PQsendQuery. Talking about more details, PQexec and PQsendQuery is designed to handle only "Q" messsage out of PostgreSQL frontend/backend protocol, while to access large objects, you need to handle "V" message. > Though I cannot guarantee anything, I may be interested into working out > a patch, if no one is already doing the same (of course I understand > that this patch wouldn't be for 9.3, which is already in its late > release cycle). > > Do you think this may be of interest? Yes, I understand your pain, and I myself think we need new APIs for large objects. Probably that would be not terribly hard. One idea would be inventing an asynchronous version of PQfn and let lo_read/lo_write allow to use the new API. -- Tatsuo Ishii SRA OSS, Inc. Japan English: http://www.sraoss.co.jp/index_en.php Japanese: http://www.sraoss.co.jp
<div dir="ltr"><br /><div class="gmail_extra"><br /><br /><div class="gmail_quote">2013/6/6 Tatsuo Ishii <span dir="ltr"><<ahref="mailto:ishii@postgresql.org" target="_blank">ishii@postgresql.org</a>></span><br /><blockquote class="gmail_quote"style="margin:0px 0px 0px 0.8ex;border-left-width:1px;border-left-color:rgb(204,204,204);border-left-style:solid;padding-left:1ex"><div class="im">>Hi.<br /> ><br /> > At the moment libpq doesn't seem to support asynchronous and<br /> > non-blockingsupport for large objects, in the style of<br /> > PQsendQuery/PQgetResult. This makes large objects hardlysuited for<br /> > single-threaded programs based on some variant of select().<br /> ><br /> > I would liketo know whether this is a deliberate decision or it is<br /> > considered a bug, and, in case, whether it is scheduledto be fixed.<br /><br /></div>Certainly not bug, since the doc clearly stats that PQsendQuery can<br /> only beused as a substituation of PQexec. (see "Asynchronous Command<br /> Processing" section" for more details). The largeobject API is<br /> completely different from PQexec and its friends, so it cannot be used<br /> with PQsendQuery.<br/><br /> Talking about more details, PQexec and PQsendQuery is designed to<br /> handle only "Q" messsageout of PostgreSQL frontend/backend protocol,<br /> while to access large objects, you need to handle "V" message.<br/></blockquote><div style="style">Really? I've specialized a C++ standard std::streambuf class by using</div><divstyle="style">only extended query protocol (by using prepared statements via PQsendPrepare,</div><div style="style">PQsendQueryPrepared)to call SQL functions like loread(), lowrite(), lo_tell(), etc.</div><div style="style">Allthese functions just needs to be called inside BEGIN block. And yes,</div><div style="style">it can be doneasynchronously.</div><blockquote class="gmail_quote" style="margin:0px 0px 0px 0.8ex;border-left-width:1px;border-left-color:rgb(204,204,204);border-left-style:solid;padding-left:1ex"><div class="im"><br/> > Though I cannot guarantee anything, I may be interested into working out<br /> > a patch, if noone is already doing the same (of course I understand<br /> > that this patch wouldn't be for 9.3, which is alreadyin its late<br /> > release cycle).<br /> ><br /> > Do you think this may be of interest?<br /><br /></div>Yes,I understand your pain, and I myself think we need new APIs for<br /> large objects. Probably that would be notterribly hard. One idea<br /> would be inventing an asynchronous version of PQfn and let<br /> lo_read/lo_write allowto use the new API.<br /></blockquote><div style="style">Yes, but according to <a href="http://www.postgresql.org/docs/9.2/static/protocol-flow.html#AEN95330">http://www.postgresql.org/docs/9.2/static/protocol-flow.html#AEN95330</a></div><div style="style">and/or <a href="http://www.postgresql.org/docs/9.2/static/libpq-fastpath.html">http://www.postgresql.org/docs/9.2/static/libpq-fastpath.html</a></div><div style="style">functioncall sub-protocol is obsolete. Thats why personally I decided to</div><div style="style">use preparedstatements.</div></div><br />-- <br />// Dmitriy.<br /><br /></div></div>
Hi. Il 05/06/2013 22:52, Dmitriy Igrishin ha scritto: >> At the moment libpq doesn't seem to support asynchronous and >> non-blocking support for large objects, in the style of >> PQsendQuery/PQgetResult. This makes large objects hardly suited for >> single-threaded programs based on some variant of select(). >> > According to http://www.postgresql.org/docs/9.2/static/lo-funcs.html > "There are server-side functions callable from SQL that correspond to each > of > the client-side functions". Hence, you can call these functions by using > asynchronous API. Thanks, I'll try this way (BTW, it may help to specify on the documentation that lo_read and lo_write lose the "_"). I wonder whether having to escape all the content for lowrite can't have a negative impact on performances. It shouldn't be too bad for my case, though. Giovanni. -- Giovanni Mascellani <mascellani@poisson.phc.unipi.it> Pisa, Italy Web: http://poisson.phc.unipi.it/~mascellani Jabber: g.mascellani@jabber.org / giovanni@elabor.homelinux.org
2013/6/8 Giovanni Mascellani <g.mascellani@gmail.com>
Hi.
Il 05/06/2013 22:52, Dmitriy Igrishin ha scritto:>> At the moment libpq doesn't seem to support asynchronous andThanks, I'll try this way (BTW, it may help to specify on the
>> non-blocking support for large objects, in the style of
>> PQsendQuery/PQgetResult. This makes large objects hardly suited for
>> single-threaded programs based on some variant of select().
>>
> According to http://www.postgresql.org/docs/9.2/static/lo-funcs.html
> "There are server-side functions callable from SQL that correspond to each
> of
> the client-side functions". Hence, you can call these functions by using
> asynchronous API.
documentation that lo_read and lo_write lose the "_"). I wonder whether
having to escape all the content for lowrite can't have a negative
impact on performances. It shouldn't be too bad for my case, though.
You may avoid escaping bytea data by using PQsendPrepare,
PQsendQueryPrepared specifying binary data format.
// Dmitriy.
> 2013/6/6 Tatsuo Ishii <ishii@postgresql.org> > >> > Hi. >> > >> > At the moment libpq doesn't seem to support asynchronous and >> > non-blocking support for large objects, in the style of >> > PQsendQuery/PQgetResult. This makes large objects hardly suited for >> > single-threaded programs based on some variant of select(). >> > >> > I would like to know whether this is a deliberate decision or it is >> > considered a bug, and, in case, whether it is scheduled to be fixed. >> >> Certainly not bug, since the doc clearly stats that PQsendQuery can >> only be used as a substituation of PQexec. (see "Asynchronous Command >> Processing" section" for more details). The large object API is >> completely different from PQexec and its friends, so it cannot be used >> with PQsendQuery. >> >> Talking about more details, PQexec and PQsendQuery is designed to >> handle only "Q" messsage out of PostgreSQL frontend/backend protocol, >> while to access large objects, you need to handle "V" message. >> > Really? I've specialized a C++ standard std::streambuf class by using > only extended query protocol (by using prepared statements via > PQsendPrepare, > PQsendQueryPrepared) to call SQL functions like loread(), lowrite(), > lo_tell(), etc. > All these functions just needs to be called inside BEGIN block. And yes, > it can be done asynchronously. Thanks for reminding me. I totally forgot about them. >> > Though I cannot guarantee anything, I may be interested into working out >> > a patch, if no one is already doing the same (of course I understand >> > that this patch wouldn't be for 9.3, which is already in its late >> > release cycle). >> > >> > Do you think this may be of interest? >> >> Yes, I understand your pain, and I myself think we need new APIs for >> large objects. Probably that would be not terribly hard. One idea >> would be inventing an asynchronous version of PQfn and let >> lo_read/lo_write allow to use the new API. >> > Yes, but according to > http://www.postgresql.org/docs/9.2/static/protocol-flow.html#AEN95330 > and/or http://www.postgresql.org/docs/9.2/static/libpq-fastpath.html > function call sub-protocol is obsolete. Thats why personally I decided to > use prepared statements. I'm not totally pleased with the comment in the doc. For me the only reason why those extended protocol functions are recommended is the binary protocol can be used. The price is parsing, planning, and preparing the query, all of them are essentially unnecessary for a large object access use case. -- Tatsuo Ishii SRA OSS, Inc. Japan English: http://www.sraoss.co.jp/index_en.php Japanese: http://www.sraoss.co.jp