Thread: Issues with building snap packages and psql
Hello, CMD has been working on building snap packages. It has been an adventure and we are very close to having them complete. In fact they are complete enough that we have published them in the snap repo. jd@jd-wks:~$ snap find postgres Name Version Developer Notes Summary postgresql96 9.6.0 cmd - PostgreSQL is a powerful, open source object-relational database system. postgresql95 9.5.3 cmd - PostgreSQL is a powerful, open source object-relational database system. postgresql94 9.4.9 cmd - PostgreSQL is a powerful, open source object-relational database system. postgresql93 9.3.14 cmd - PostgreSQL is a powerful, open source object-relational database system. However, we have found a problem with permissions that we haven't been able to overcome. The error presents as a simple: could not save history to file "/home/jd/.psql_history": Permission denied We tried the work around as presented here: https://www.commandprompt.com/blog/psql_tip_change_history_location/ It didn't work and I couldn't figure out why. It was especially confusing because if you are in psql via snap $HOME is set correctly: postgres=# \! export export HOME='/home/jd/snap/postgresql96/1' Then I traced it and saw this: open("/etc/passwd", O_RDONLY|O_CLOEXEC) = 7 lseek(7, 0, SEEK_CUR) = 0 fstat(7, {st_mode=S_IFREG|0644, st_size=2581, ...}) = 0 mmap(NULL, 2581, PROT_READ, MAP_SHARED, 7, 0) = 0x7f1f9e5fc000 lseek(7, 2581, SEEK_SET) = 2581 munmap(0x7f1f9e5fc000, 2581) = 0 close(7) = 0 [...] futex(0x1096088, FUTEX_WAIT, 0, NULL) = 0 futex(0x1096088, FUTEX_WAIT, 0, NULL/home/jd/.psqlrc: Permission denied psql (9.6.0) Type "help" for help. postgres=# \q could not save history to file "/home/jd/.psql_history": Permission denied <unfinished ...> +++ exited with 0 +++ Is psql getting the home directory from /etc/passwd? If so what can we do about that? Sincerely, JD -- Command Prompt, Inc. http://the.postgres.company/ +1-503-667-4564 PostgreSQL Centered full stack support, consulting and development. Everyone appreciates your honesty, until you are honest with them. Unless otherwise stated, opinions are my own.
"Joshua D. Drake" <jd@commandprompt.com> writes:
> Is psql getting the home directory from /etc/passwd?
Yes, of course.
> If so what can we do about that?
Fix your /etc/passwd entry? Seems unlikely that psql is the only
program that believes that entry.
regards, tom lane
On 10/26/2016 10:16 AM, Tom Lane wrote: > "Joshua D. Drake" <jd@commandprompt.com> writes: >> Is psql getting the home directory from /etc/passwd? > > Yes, of course. > >> If so what can we do about that? > > Fix your /etc/passwd entry? Seems unlikely that psql is the only > program that believes that entry. It isn't that simple. Snap packages are similar to a chroot. My home directory *is* /home/jd but snap packages can't write to that because of the isolation. Sincerely, JD -- Command Prompt, Inc. http://the.postgres.company/ +1-503-667-4564 PostgreSQL Centered full stack support, consulting and development. Everyone appreciates your honesty, until you are honest with them. Unless otherwise stated, opinions are my own.
On October 26, 2016 8:09:11 PM GMT+03:00, "Joshua D. Drake" <jd@commandprompt.com> wrote: >postgres=# \! export >export HOME='/home/jd/snap/postgresql96/1' That doesn't do much. Isn't that just spawning a shell in which you set the variable, which then immediately exits? And thisthe value isn't set anywhere relevant? Andres -- Sent from my Android device with K-9 Mail. Please excuse my brevity.
On 10/26/2016 10:26 AM, Andres Freund wrote: > > > On October 26, 2016 8:09:11 PM GMT+03:00, "Joshua D. Drake" <jd@commandprompt.com> wrote: > >> postgres=# \! export >> export HOME='/home/jd/snap/postgresql96/1' > > That doesn't do much. Isn't that just spawning a shell in which you set the variable, which then immediately exits? Andthis the value isn't set anywhere relevant? I was assuming (and TGL corrected me) that psql was going to pull the $HOME variable to determine where to read/write .psqlrc and .psql_history. Sincerely, JD -- Command Prompt, Inc. http://the.postgres.company/ +1-503-667-4564 PostgreSQL Centered full stack support, consulting and development. Everyone appreciates your honesty, until you are honest with them. Unless otherwise stated, opinions are my own.
Hello, After some further investigation and collaboration with #postgresql it looks like what we have here is an actual bug or at least improper implementation. Apparently, we use getpwuid on the euid to locate the 'home directory'. This is incorrect (as well as inconsistent to our own documentation) [1]. Our documentation says that dotfiles are ~/.whatever, and ~ on its own in shell means the value of $HOME, it does NOT mean the current user's homedir from getpwuid. It is also inconsistent with how other programs behave. For example if psql uses readline, and you change the value of $HOME, then readline uses $HOME/.inputrc whereas psql does not use $HOME/.psqlrc. Sincerely, JD 1. http://pubs.opengroup.org/onlinepubs/9699919799/utilities/V3_chap02.html#tag_18_06_01 -- Command Prompt, Inc. http://the.postgres.company/ +1-503-667-4564 PostgreSQL Centered full stack support, consulting and development. Everyone appreciates your honesty, until you are honest with them. Unless otherwise stated, opinions are my own.
On Wed, Oct 26, 2016 at 2:22 PM, Joshua D. Drake <jd@commandprompt.com> wrote: > After some further investigation and collaboration with #postgresql it looks > like what we have here is an actual bug or at least improper implementation. > Apparently, we use getpwuid on the euid to locate the 'home directory'. This > is incorrect (as well as inconsistent to our own documentation) [1]. Our > documentation says that dotfiles are ~/.whatever, and ~ on its own in shell > means the value of $HOME, it does NOT mean the current user's homedir from > getpwuid. > > > It is also inconsistent with how other programs behave. For example if psql > uses readline, and you change the value of $HOME, then readline uses > $HOME/.inputrc whereas psql does not use $HOME/.psqlrc. Sounds like the same issue that Rudolf Gavlas complained about in https://www.postgresql.org/message-id/CAEH6cQqbdbXoUHJBbX9ixwfjFFsUC-a8hFntKcci=diWgBb3fQ@mail.gmail.com -- Robert Haas EnterpriseDB: http://www.enterprisedb.com The Enterprise PostgreSQL Company
On 10/26/2016 11:33 AM, Robert Haas wrote: > On Wed, Oct 26, 2016 at 2:22 PM, Joshua D. Drake <jd@commandprompt.com> wrote: >> After some further investigation and collaboration with #postgresql it looks >> like what we have here is an actual bug or at least improper implementation. >> Apparently, we use getpwuid on the euid to locate the 'home directory'. This >> is incorrect (as well as inconsistent to our own documentation) [1]. Our >> documentation says that dotfiles are ~/.whatever, and ~ on its own in shell >> means the value of $HOME, it does NOT mean the current user's homedir from >> getpwuid. >> >> >> It is also inconsistent with how other programs behave. For example if psql >> uses readline, and you change the value of $HOME, then readline uses >> $HOME/.inputrc whereas psql does not use $HOME/.psqlrc. > > Sounds like the same issue that Rudolf Gavlas complained about in > https://www.postgresql.org/message-id/CAEH6cQqbdbXoUHJBbX9ixwfjFFsUC-a8hFntKcci=diWgBb3fQ@mail.gmail.com It does sound similar, unfortunately the limited response doesn't seem to address the issue in that we are doing it in a way that nobody (or at least commonly) else does. To me this sounds strikingly familiar to how we ignored how everyone else handled commenting out parameters vs how we thought we were right until we finally decided everyone else was already correct and changed it. (metaphorically speaking) That said, I am not interested in a philosophical argument. It appears to me that based on the referenced PDF that we are wrong. How do we fix it? What are the implications of that? I just want to solve the problem. Sincerely, JD -- Command Prompt, Inc. http://the.postgres.company/ +1-503-667-4564 PostgreSQL Centered full stack support, consulting and development. Everyone appreciates your honesty, until you are honest with them. Unless otherwise stated, opinions are my own.
On 10/26/2016 12:10 PM, Joshua D. Drake wrote: > On 10/26/2016 11:33 AM, Robert Haas wrote: >> On Wed, Oct 26, 2016 at 2:22 PM, Joshua D. Drake >>> It is also inconsistent with how other programs behave. For example >>> if psql >>> uses readline, and you change the value of $HOME, then readline uses >>> $HOME/.inputrc whereas psql does not use $HOME/.psqlrc. >> >> Sounds like the same issue that Rudolf Gavlas complained about in >> https://www.postgresql.org/message-id/CAEH6cQqbdbXoUHJBbX9ixwfjFFsUC-a8hFntKcci=diWgBb3fQ@mail.gmail.com This problem clearly an improper implementation on our part. We have found a work around by wrapping the launch of psql so we can set PSQLRC. At some point, perhaps with v10 we could change this. This problem has been noted on other ends of the innertubes as well (we submitted this as a potential bug to Ubuntu and they too found that we don't handle this correctly). Sincerely, JD -- Command Prompt, Inc. http://the.postgres.company/ +1-503-667-4564 PostgreSQL Centered full stack support, consulting and development. Everyone appreciates your honesty, until you are honest with them. Unless otherwise stated, opinions are my own.
On Thu, Oct 27, 2016 at 12:24 PM, Joshua D. Drake <jd@commandprompt.com> wrote: > This problem clearly an improper implementation on our part. We have found a > work around by wrapping the launch of psql so we can set PSQLRC. At some > point, perhaps with v10 we could change this. This problem has been noted on > other ends of the innertubes as well (we submitted this as a potential bug > to Ubuntu and they too found that we don't handle this correctly). Yes, I think the evidence is growing that we ought to change this. -- Robert Haas EnterpriseDB: http://www.enterprisedb.com The Enterprise PostgreSQL Company