Thread: Make ON_ERROR_STOP stop on shell script failure

Hi,

"""\set ON_ERROR_STOP on""" stops any subsequent incoming query that 
comes after an error of an SQL, but does not stop after a shell script 
ran by """\! <some command>""" returning values other than 0, -1, or 
127, which suggests a failure in the result of the shell script.

For example, suppose that below is an SQL file.
\set ON_ERROR_STOP on
SELECT 1;
\! false
SELECT 2;

The current design allows SELECT 2 even though the shell script returns 
a value indicating a failure.

I thought that this action is rather unexpected since, based on the word 
"""ON_ERROR_STOP""", ones may expect that failures of shell scripts 
should halt the incoming instructions as well.
One clear solution is to let failures of shell script stop incoming 
queries just like how errors of SQLs do currently. Thoughts?

Tatsu

At Fri, 16 Sep 2022 15:55:33 +0900, bt22nakamorit <bt22nakamorit@oss.nttdata.com> wrote in 
> Hi,
> 
> """\set ON_ERROR_STOP on""" stops any subsequent incoming query that
> comes after an error of an SQL, but does not stop after a shell script
> ran by """\! <some command>""" returning values other than 0, -1, or
> 127, which suggests a failure in the result of the shell script.
> 
> For example, suppose that below is an SQL file.
> \set ON_ERROR_STOP on
> SELECT 1;
> \! false
> SELECT 2;
> 
> The current design allows SELECT 2 even though the shell script
> returns a value indicating a failure.

Since the "false" command did not "error out"?

> I thought that this action is rather unexpected since, based on the
> word """ON_ERROR_STOP""", ones may expect that failures of shell
> scripts should halt the incoming instructions as well.
> One clear solution is to let failures of shell script stop incoming
> queries just like how errors of SQLs do currently. Thoughts?

I'm not sure we want to regard any exit status from a succssful run as
a failure.

On the other hand, the proposed behavior seems useful to me.

So +1 from me to the proposal, assuming the corresponding edit of the
documentation happens.

regards.

-- 
Kyotaro Horiguchi
NTT Open Source Software Center

2022-09-16 17:30 に Kyotaro Horiguchi さんは書きました:
> At Fri, 16 Sep 2022 15:55:33 +0900, bt22nakamorit
> <bt22nakamorit@oss.nttdata.com> wrote in
>> Hi,
>> 
>> """\set ON_ERROR_STOP on""" stops any subsequent incoming query that
>> comes after an error of an SQL, but does not stop after a shell script
>> ran by """\! <some command>""" returning values other than 0, -1, or
>> 127, which suggests a failure in the result of the shell script.
>> 
>> For example, suppose that below is an SQL file.
>> \set ON_ERROR_STOP on
>> SELECT 1;
>> \! false
>> SELECT 2;
>> 
>> The current design allows SELECT 2 even though the shell script
>> returns a value indicating a failure.
> 
> Since the "false" command did not "error out"?
> 
>> I thought that this action is rather unexpected since, based on the
>> word """ON_ERROR_STOP""", ones may expect that failures of shell
>> scripts should halt the incoming instructions as well.
>> One clear solution is to let failures of shell script stop incoming
>> queries just like how errors of SQLs do currently. Thoughts?
> 
> I'm not sure we want to regard any exit status from a succssful run as
> a failure.
> 
> On the other hand, the proposed behavior seems useful to me.
> 
> So +1 from me to the proposal, assuming the corresponding edit of the
> documentation happens.
> 
> regards.

> Since the "false" command did not "error out"?
"false" command returns 1 which is an exit status code that indicates 
failure, but not error.
I think it does not "error out" if that is what you mean.

> So +1 from me to the proposal, assuming the corresponding edit of the
> documentation happens.
I will work on editing the document and share further updates.

Thank you!
Tatsu

2022-09-17 09:44 に bt22nakamorit さんは書きました:
> 2022-09-16 17:30 に Kyotaro Horiguchi さんは書きました:
>> At Fri, 16 Sep 2022 15:55:33 +0900, bt22nakamorit
>> <bt22nakamorit@oss.nttdata.com> wrote in
>>> Hi,
>>> 
>>> """\set ON_ERROR_STOP on""" stops any subsequent incoming query that
>>> comes after an error of an SQL, but does not stop after a shell 
>>> script
>>> ran by """\! <some command>""" returning values other than 0, -1, or
>>> 127, which suggests a failure in the result of the shell script.
>>> 
>>> For example, suppose that below is an SQL file.
>>> \set ON_ERROR_STOP on
>>> SELECT 1;
>>> \! false
>>> SELECT 2;
>>> 
>>> The current design allows SELECT 2 even though the shell script
>>> returns a value indicating a failure.
>> 
>> Since the "false" command did not "error out"?
>> 
>>> I thought that this action is rather unexpected since, based on the
>>> word """ON_ERROR_STOP""", ones may expect that failures of shell
>>> scripts should halt the incoming instructions as well.
>>> One clear solution is to let failures of shell script stop incoming
>>> queries just like how errors of SQLs do currently. Thoughts?
>> 
>> I'm not sure we want to regard any exit status from a succssful run as
>> a failure.
>> 
>> On the other hand, the proposed behavior seems useful to me.
>> 
>> So +1 from me to the proposal, assuming the corresponding edit of the
>> documentation happens.
>> 
>> regards.
> 
>> Since the "false" command did not "error out"?
> "false" command returns 1 which is an exit status code that indicates
> failure, but not error.
> I think it does not "error out" if that is what you mean.
> 
>> So +1 from me to the proposal, assuming the corresponding edit of the
>> documentation happens.
> I will work on editing the document and share further updates.
> 
> Thank you!
> Tatsu

I edited the documentation for ON_ERROR_STOP.
Any other suggestions?

Tatsu

On 2022/09/20 15:15, bt22nakamorit wrote:
>>>> I thought that this action is rather unexpected since, based on the
>>>> word """ON_ERROR_STOP""", ones may expect that failures of shell
>>>> scripts should halt the incoming instructions as well.
>>>> One clear solution is to let failures of shell script stop incoming
>>>> queries just like how errors of SQLs do currently. Thoughts?

+1


> I edited the documentation for ON_ERROR_STOP.
> Any other suggestions?

Thanks for the patch!
Could you add it to the next CommitFest so that we don't forget it?


We can execute the shell commands via psql in various ways
other than \! meta-command. For example,

* `command`
* \g | command
* \gx | command
* \o | command
* \w | command
* \copy ... program 'command'

ON_ERROR_STOP should handle not only \! but also all the above in the same way?


One concern about this patch is that some applications already depend on
the current behavior of ON_ERROR_STOP, i.e., psql doesn't stop even when
the shell command returns non-zero exit code. If so, we might need to
extend ON_ERROR_STOP so that it accepts the following setting values.

* off - don't stop even when either sql or shell fails (same as the current behavior)
* on or sql - stop only whensql fails (same as the current behavior)
* shell - stop only when shell fails
* all - stop when either sql or shell fails

Thought?

Regards,

-- 
Fujii Masao
Advanced Computing Technology Center
Research and Development Headquarters
NTT DATA CORPORATION

At Wed, 21 Sep 2022 11:45:07 +0900, Fujii Masao <masao.fujii@oss.nttdata.com> wrote in 
> 
> 
> On 2022/09/20 15:15, bt22nakamorit wrote:
> >>>> I thought that this action is rather unexpected since, based on the
> >>>> word """ON_ERROR_STOP""", ones may expect that failures of shell
> >>>> scripts should halt the incoming instructions as well.
> >>>> One clear solution is to let failures of shell script stop incoming
> >>>> queries just like how errors of SQLs do currently. Thoughts?
> 
> +1
> 
> 
> > I edited the documentation for ON_ERROR_STOP.
> > Any other suggestions?
> 
> Thanks for the patch!
> Could you add it to the next CommitFest so that we don't forget it?
> 
> 
> We can execute the shell commands via psql in various ways
> other than \! meta-command. For example,
> 
> * `command`
> * \g | command
> * \gx | command
> * \o | command
> * \w | command
> * \copy ... program 'command'
> 
> ON_ERROR_STOP should handle not only \! but also all the above in the
> same way?

+1

> One concern about this patch is that some applications already depend
> on
> the current behavior of ON_ERROR_STOP, i.e., psql doesn't stop even
> when
> the shell command returns non-zero exit code. If so, we might need to
> extend ON_ERROR_STOP so that it accepts the following setting values.
> 
> * off - don't stop even when either sql or shell fails (same as the
> * current behavior)
> * on or sql - stop only whensql fails (same as the current behavior)
> * shell - stop only when shell fails
> * all - stop when either sql or shell fails
> 
> Thought?

+1

regards.

-- 
Kyotaro Horiguchi
NTT Open Source Software Center

Fujii Masao:
> One concern about this patch is that some applications already depend on
> the current behavior of ON_ERROR_STOP, i.e., psql doesn't stop even when
> the shell command returns non-zero exit code. If so, we might need to
> extend ON_ERROR_STOP so that it accepts the following setting values.

I just got bitten by this and I definitely consider this a bug. I expect 
psql to stop when a shell script fails and I have ON_ERROR_STOP set. I 
don't think this should be made more complicated with different settings.

If someone needs to have ON_ERROR_STOP set, but continue execution after 
a certain shell command, they could still do something like this:

\! might_fail || true

Best

Wolfgang

On 2022-09-20 15:15, bt22nakamorit wrote:

> I edited the documentation for ON_ERROR_STOP.
> Any other suggestions?

Thanks for the patch!

>    if (result == 127 || result == -1)
>    {
>        pg_log_error("\\!: failed");
>        return false;
>    }
>    else if (result != 0) {
>        pg_log_error("command failed");
>        return false;

Since it would be hard to understand the cause of failures from these 
two messages,　it might be better to clarify them in the messages.

The former comes from failures of child process creation or execution on 
it and the latter occurs when child process creation and execution 
succeeded but the return code is not 0, doesn't it?


I also felt it'd be natural that the latter message also begins with 
"\\!" since both message concerns with \!.

How do you think?

-- 
Regards,

--
Atsushi Torikoshi
NTT DATA CORPORATION

2022-09-28 21:49 に torikoshia さんは書きました:
>>    if (result == 127 || result == -1)
>>    {
>>        pg_log_error("\\!: failed");
>>        return false;
>>    }
>>    else if (result != 0) {
>>        pg_log_error("command failed");
>>        return false;
> 
> Since it would be hard to understand the cause of failures from these
> two messages,　it might be better to clarify them in the messages.
> 
> The former comes from failures of child process creation or execution
> on it and the latter occurs when child process creation and execution
> succeeded but the return code is not 0, doesn't it?
> 
> 
> I also felt it'd be natural that the latter message also begins with
> "\\!" since both message concerns with \!.
> 
> How do you think?

Thank you for the feedback!
I agree that the messages should be more clear.
\\!: command was not executed
\\!: command failed
Would these two messages be enough to describe the two cases?

Tatsu

At Thu, 29 Sep 2022 11:29:40 +0900, bt22nakamorit <bt22nakamorit@oss.nttdata.com> wrote in 
> 2022-09-28 21:49 に torikoshia さんは書きました:
> >>    if (result == 127 || result == -1)
> >>    {
> >>        pg_log_error("\\!: failed");
> >>        return false;
> >>    }
> >>    else if (result != 0) {
> >>        pg_log_error("command failed");
> >>        return false;
> > Since it would be hard to understand the cause of failures from these
> > two messages,　it might be better to clarify them in the messages.
> > The former comes from failures of child process creation or execution
> > on it and the latter occurs when child process creation and execution
> > succeeded but the return code is not 0, doesn't it?
> > I also felt it'd be natural that the latter message also begins with
> > "\\!" since both message concerns with \!.
> > How do you think?
> 
> Thank you for the feedback!
> I agree that the messages should be more clear.
> \\!: command was not executed
> \\!: command failed
> Would these two messages be enough to describe the two cases?

FWIW, I would spell these as something like this:

> \\!: command execution failure: %m
> \\!: command returned failure status: %d

regards.

-- 
Kyotaro Horiguchi
NTT Open Source Software Center

At Thu, 29 Sep 2022 12:35:04 +0900 (JST), Kyotaro Horiguchi <horikyota.ntt@gmail.com> wrote in 
> > Thank you for the feedback!
> > I agree that the messages should be more clear.
> > \\!: command was not executed
> > \\!: command failed
> > Would these two messages be enough to describe the two cases?
> 
> FWIW, I would spell these as something like this:
> 
> > \\!: command execution failure: %m

The following might be more complient to our policy.

> \\!: failed to execute command \"%s\": %m


> > \\!: command returned failure status: %d

regards.

-- 
Kyotaro Horiguchi
NTT Open Source Software Center

2022-09-21 11:45 に Fujii Masao wrote:
> We can execute the shell commands via psql in various ways
> other than \! meta-command. For example,
> 
> * `command`
> * \g | command
> * \gx | command
> * \o | command
> * \w | command
> * \copy ... program 'command'
> 
> ON_ERROR_STOP should handle not only \! but also all the above in the 
> same way?
> 
> 
> One concern about this patch is that some applications already depend 
> on
> the current behavior of ON_ERROR_STOP, i.e., psql doesn't stop even 
> when
> the shell command returns non-zero exit code. If so, we might need to
> extend ON_ERROR_STOP so that it accepts the following setting values.
> 
> * off - don't stop even when either sql or shell fails (same as the
> current behavior)
> * on or sql - stop only whensql fails (same as the current behavior)
> * shell - stop only when shell fails
> * all - stop when either sql or shell fails
> 
> Thought?
> 
> Regards,

I agree that some applications may depend on the behavior of previous 
ON_ERROR_STOP.
I created a patch that implements off, on, shell, and all option for 
ON_ERROR_STOP.
I also edited the code for \g, \o, \w, and \set in addition to \! to 
return exit status of shell commands for ON_ERROR_STOP.

There were discussions regarding the error messages for when shell 
command fails.
I have found that \copy already handles exit status of shell commands 
when it executes one, so I copied the messages from there.
More specifically, I referred to """bool do_copy(const char *args)""" in 
src/bin/psql/copy.c

Any feedback would be very much appreciated.

Tatsu

On 2022/09/30 16:54, bt22nakamorit wrote:
> 2022-09-21 11:45 に Fujii Masao wrote:
>> We can execute the shell commands via psql in various ways
>> other than \! meta-command. For example,
>>
>> * `command`
>> * \g | command
>> * \gx | command
>> * \o | command
>> * \w | command
>> * \copy ... program 'command'
>>
>> ON_ERROR_STOP should handle not only \! but also all the above in the same way?
>>
>>
>> One concern about this patch is that some applications already depend on
>> the current behavior of ON_ERROR_STOP, i.e., psql doesn't stop even when
>> the shell command returns non-zero exit code. If so, we might need to
>> extend ON_ERROR_STOP so that it accepts the following setting values.
>>
>> * off - don't stop even when either sql or shell fails (same as the
>> current behavior)
>> * on or sql - stop only whensql fails (same as the current behavior)
>> * shell - stop only when shell fails
>> * all - stop when either sql or shell fails
>>
>> Thought?
>>
>> Regards,
> 
> I agree that some applications may depend on the behavior of previous ON_ERROR_STOP.
> I created a patch that implements off, on, shell, and all option for ON_ERROR_STOP.
> I also edited the code for \g, \o, \w, and \set in addition to \! to return exit status of shell commands for
ON_ERROR_STOP.
> 
> There were discussions regarding the error messages for when shell command fails.
> I have found that \copy already handles exit status of shell commands when it executes one, so I copied the messages
fromthere.
 
> More specifically, I referred to """bool do_copy(const char *args)""" in src/bin/psql/copy.c
> 
> Any feedback would be very much appreciated.

Thanks for updating the patch!

The patch failed to be applied into the master cleanly. Could you rebase it?

patching file src/bin/psql/common.c
Hunk #1 succeeded at 94 (offset 4 lines).
Hunk #2 succeeded at 104 (offset 4 lines).
Hunk #3 succeeded at 133 (offset 4 lines).
Hunk #4 succeeded at 1869 with fuzz 1 (offset 1162 lines).
Hunk #5 FAILED at 2624.
1 out of 5 hunks FAILED -- saving rejects to file src/bin/psql/common.c.rej

Regards,

-- 
Fujii Masao
Advanced Computing Technology Center
Research and Development Headquarters
NTT DATA CORPORATION

2022-10-07 17:16 Fujii Masao wrote:
> The patch failed to be applied into the master cleanly. Could you 
> rebase it?
> 
> patching file src/bin/psql/common.c
> Hunk #1 succeeded at 94 (offset 4 lines).
> Hunk #2 succeeded at 104 (offset 4 lines).
> Hunk #3 succeeded at 133 (offset 4 lines).
> Hunk #4 succeeded at 1869 with fuzz 1 (offset 1162 lines).
> Hunk #5 FAILED at 2624.
> 1 out of 5 hunks FAILED -- saving rejects to file 
> src/bin/psql/common.c.rej

Thank you for checking.
I edited the patch so that it would apply to the latest master branch.
Please mention if there are any other problems.

Best,
Tatsuhiro Nakamori

There was a mistake in the error message for \! so I updated the patch.

Best,
Tatsuhiro Nakamori

On Fri, Sep 16, 2022 at 03:55:33PM +0900, bt22nakamorit wrote:
> Hi,
> 
> """\set ON_ERROR_STOP on""" stops any subsequent incoming query that comes
> after an error of an SQL, but does not stop after a shell script ran by
> """\! <some command>""" returning values other than 0, -1, or 127, which
> suggests a failure in the result of the shell script.

Actually, I think this could be described as a wider problem (not just
ON_ERROR_STOP).  The shell's exit status is being ignored (except for -1
and 127).

Shouldn't the user be able to do something with the exit status ?
Right now, it seems like they'd need to wrap the shellscript with
"if ! ...; then echo failed; fi"
and then \gset and compare with "failed"

I think it'd be a lot better to expose the script status to psql.
(without having to write "foo; echo status=$?").

Another consideration is that shellscripts can exit with a nonzero
status due to the most recent conditional (like: if || &&).

For example, consider shell command like:
"if foo; then bar; fi" or "foo && bar"

If foo has nonzero status, then bar isn't run.

If that's the entire shell script, the shell will *also* exit with foo's
nonzero status.  (That's the reason why people write "exit 0" as the
last line of a shell script.  It's easy to believe that it was going to
"exit 0" in any case; but, what it was actually going to do was to "exit
$?", and $? can be nonzero after conditionals, even in "set -e" mode).

So a psql script like this would start to report as a failure any time
"foo" was false, even if that's the normal/typical case.

-- 
Justin

------- Original Message -------
On Tuesday, November 22nd, 2022 at 20:10, bt22nakamorit <bt22nakamorit@oss.nttdata.com> wrote:


> There was a mistake in the error message for \! so I updated the patch.
>
> Best,
> Tatsuhiro Nakamori

Hi

I was checking your patch and seems that it failed to be applied into the
master cleanly. Could you please rebase it?

--
Matheus Alcantara

On 2022-10-12 2:13 a.m., bt22nakamorit wrote:
> There was a mistake in the error message for \! so I updated the patch.
>
Tried to apply this patch to the master branch, but got the error like 
below.
$ git apply --check patch-view.diff
error: patch failed: src/bin/psql/command.c:2693
error: src/bin/psql/command.c: patch does not apply

I think there are some tests related with "ON_ERROR_STOP" in 
src/bin/psql/t/001_basic.pl, and we should consider to add corresponding 
test cases for "on/off/shell/all" to this patch.


Best regards,

David

On 23/11/2022 00:16, Matheus Alcantara wrote:
> ------- Original Message -------
> On Tuesday, November 22nd, 2022 at 20:10, bt22nakamorit <bt22nakamorit@oss.nttdata.com> wrote:
>
>
>> There was a mistake in the error message for \! so I updated the patch.
>>
>> Best,
>> Tatsuhiro Nakamori
> Hi
>
> I was checking your patch and seems that it failed to be applied into the
> master cleanly. Could you please rebase it?

Was also looking into this patch.

Tatsuhiro: can you please send a rebased version?


Thanks

-- 
                Andreas 'ads' Scherbaum
German PostgreSQL User Group
European PostgreSQL User Group - Board of Directors
Volunteer Regional Contact, Germany - PostgreSQL Project

On 16/02/2023 20:33, Andreas 'ads' Scherbaum wrote:
> On 23/11/2022 00:16, Matheus Alcantara wrote:
>> ------- Original Message -------
>> On Tuesday, November 22nd, 2022 at 20:10, bt22nakamorit 
>> <bt22nakamorit@oss.nttdata.com> wrote:
>>
>>
>>> There was a mistake in the error message for \! so I updated the patch.
>>>
>>> Best,
>>> Tatsuhiro Nakamori
>> Hi
>>
>> I was checking your patch and seems that it failed to be applied into 
>> the
>> master cleanly. Could you please rebase it?
>
> Was also looking into this patch.
>
> Tatsuhiro: can you please send a rebased version?

The email address is bouncing. That might be why ...

-- 
                Andreas 'ads' Scherbaum
German PostgreSQL User Group
European PostgreSQL User Group - Board of Directors
Volunteer Regional Contact, Germany - PostgreSQL Project

So I took a look at this patch. The conflict is with 2fe3bdbd691
committed by Peter Eisentraut which added error checks for pipes.
Afaics the behaviour is now for pipe commands returning non-zero to
cause an error *always* regardless of the setting of ON_ERROR_STOP.

I'm not entirely sure that's sensible actually. If you write to a pipe
that ends in grep and it happens to produce no matching rows you may
actually be quite surprised when that causes your script to fail...

But if you remove that failing hunk the resulting patch does apply. I
don't see any tests so ... I don't know if the behaviour is still
sensible. A quick read gives me the impression that now it's actually
inconsistent in the other direction where it stops sometimes more
often than the user might expect.

I also don't understand the difference between ON_ERROR_STOP_ON and
ON_ERROR_STOP_ALL. Nor why we would want ON_ERROR_STOP_SHELL which
stops only on shell errors, rather than, say, ON_ERROR_STOP_SQL to do
the converse which would at least match the historical behaviour?

Pruning bouncing email address -- please respond from this point in
thread, not previous messages.


-- 
Gregory Stark
As Commitfest Manager

On Mon, 20 Mar 2023 at 14:34, Gregory Stark (as CFM)
<stark.cfm@gmail.com> wrote:
>
> Pruning bouncing email address -- please respond from this point in
> thread, not previous messages.

Oh for heaven's sake. Trying again to prune bouncing email address.
Please respond from *here* on the thread.... Sorry


-- 
Gregory Stark
As Commitfest Manager

On 20.03.23 19:31, Greg Stark wrote:
> So I took a look at this patch. The conflict is with 2fe3bdbd691
> committed by Peter Eisentraut which added error checks for pipes.
> Afaics the behaviour is now for pipe commands returning non-zero to
> cause an error*always*  regardless of the setting of ON_ERROR_STOP.
> 
> I'm not entirely sure that's sensible actually. If you write to a pipe
> that ends in grep and it happens to produce no matching rows you may
> actually be quite surprised when that causes your script to fail...

The only thing that that patch changed in psql was the \w command, and 
AFAICT, ON_ERROR_STOP is still respected:

$ cat test.sql
\w |foo

$ psql -f test.sql
sh: foo: command not found
psql:test.sql:1: error: |foo: command not found
$ echo $?
0

$ psql -f test.sql -v ON_ERROR_STOP=1
sh: foo: command not found
psql:test.sql:1: error: |foo: command not found
$ echo $?
3

This patch hasn't applied for quite some time, has been waiting on author since
December, and the thread has stalled.  I'm marking this Returned with Feedback
for now, please feel free to resubmit to a future CF when there is renewed
interest in working on this.

--
Daniel Gustafsson