Post-merge notes

We can close the issue/PR below after this PR is merged.
#9727
#9728

The INSTALL instruction should be modified because it will no longer require pre-installation process.
https://github.com/Chia-Network/chia-blockchain/wiki/INSTALL#centosred-hatfedora

@wjblanke @altendky
(This PR updates install scripts. Maybe more members should review the changed files)

Changelog

Changed

• Removed sudo npm install n from install-gui.sh and put cd build_scripts/npm_global; npm ci instead.
  • n is a nodejs version switcher
  • n is installed to build_scripts/npm_global/node_modules if npm>=7 is not available on system.
• If npm>=7 is not available, it automatically install npm>=7 to <chia-blockchain root dir>/.n.
  • So it won't override existing global npm/nodejs anymore.
  • n fetches nodejs tarball from dist server. However there seems no prebuild version for OpenBSD/FreeBSD available. So if on OpenBSD/FreeBSD and npm>=7 is not available, the script asks user to install npm>=7 manually.
• This now installs sqlite/python to CentOS7/AmazonLinux2 automatically if it is not installed.
  This greatly simplifies INSTALL process over different OSes
  because it will no longer require pre-install manual installation and cd chia-blockchain; sh install.sh will do everything for users.
• If the version of sqlite linked to Python is less than 3.8, it shows an error message and exit.
  (chia-blockchain requires sqlite>=3.8 for creating partial index)
• If user attemps to run install.sh when there is a newer version of a python than a python already installed in venv directory, install.sh now clears venv folder and installs python packages corresponding to the new python.

Fixed

• Fixed an issue where install-gui.sh failed if only npm<7 was available on the system.
• Fixed an issue where install-gui.sh did not work as expected when an user runs the script not from a directory where the shell script file exists.
• Fixed an issue where install-gui.sh tried to run sudo apt-get npm nodejs libxss1 even if nodejs is already installed.
• Fixed an issue where install.sh showed unnecessary stdout/stderr message from pacman
• Fixed an issue where there was no python install on RHEL in install.sh
• Fixed an issue where install.sh tried to run sudo apt-get install bc -y even if bc is already installed.

Added

• Added CentOS install test for CI (.github/workflow/test-install-scripts.yml)
• Added start-gui.sh.
  • This automatically tries to load local nodejs environment if it's available.
  • If npm>=7 is not available from system/local nodejs environment, asks users to install npm>=7
  • This checks whether GUI build files are available on startup. If not, tell user to build the GUI app before starting.
    • This is to help users who accidentally run start-gui.sh before running install-gui.sh.

Found out that Ubuntu18 which is ec2 instance(t2.medium) with 4GB RAM struggles to process GUI build. Will do some researching how much RAM is sufficient to build GUI.

In my observation, install-gui.sh consumes 5.1GB at peak. So when running install-gui.sh user need to have at least 5.1GB free RAM.

I tested Ubuntu 18 and Rocky OS 8. Both worked fine, although those were not an entirely new OS install (I had installed previous versions of chia)

Centos 7 doesn't work as there is apparently no python 3.9 package and this causes a failure:
No package python39 available

@emlowe Thanks for testing. CentOS7, I'll check and fix the script. IIRC CentOS7 even do not have sudo installed by default so checking whether sudo exists is what I'm going to modify. But as far as installing sudo, I think I shouldn't because such an important security module should be installed by admin of the computer and should not be urged to install by non-admin user/application.
Will do my best to fix anyway.

I don't think I've ever installed Centos 7 and not had sudo.

IIRC CentOS7 in docker container used in CI runner for the rpm installer showed error that sudo was not available. I remember I was surprised to see that. I'll install CentOS7 on my local Hyper-V environment from CD installer and see whether sudo actually exists or not by default

I've found that installing Python on CentOS7 does not come easy. You cannot simply sudo yum install python3.9 to install python3 because default package repository for CentOS7 seems not to have python(>=3.7) module(If you run sudo yum install python3 it tries to install Python3.6).
I'll find the way to install chia-compatible version of Python.

I didn't add any CentOS docker (CLI) installer tests exactly because they seemed to not provide new enough packages. Isn't CentOS dead anyways? Was Rocky, as you reference Earle, the replacement? Anyways, not a clean install here but worked straight away on Ubuntu 20.04. Thanks for all the work on this. I guess I'll go look at adding Rocky to the (CLI) installer test set.

I had earlier made the call that CentOS 8 was as far back as we need to go. I think it's safe to return to that thought.

For reference on CentOS installs. https://github.com/Chia-Network/chia-blockchain/wiki/INSTALL#centosred-hatfedora

I've updated install.sh to be run without error on CentOS7. I'm testing CentOS8 now.

Centos moved from a downstream model (Now called Centos Linux) to an upstream model and rebranded as Centos Stream

Centos Linux 7 is supported until 2024
Centos Linux 8 is EOL

Centos Stream 8 is supported until 2024, but because of the change in the model, this release gets patches continuously before they go into RHEL - there I don't think anyone is using it as they used to since it is upstream of RHEL it is no longer a bug-for-bug proxy for RHEL. Centos Stream 8 has no minor patch versions, as patches are just rolled into it whenever

Centos Stream 9 is also supported until 2027

Rocky 8 is a community-driven response to this change to make a release that is downstream of RHEL and is bug-for-bug compatible with RHEL 8 and is supported until 2029

Here endeth the saga that is Centos.

@wjblanke @emlowe @altendky
I tested the installer and 90% of work has been done.
For now I need to solve a problem where at AmazonLinux2 the default version of sqlite3 available is 3.7, which causes an error creating partial index. (create index xxx on yyy where zzz).
Once this issue is resolved, I'll suggest you to merge this PR.
Just FYI.
See #9921 (comment) for detail

Thanks for keeping this going. Whatever you have added support for in install.sh should get added to the matrix in test-install-scripts.yml.

@altendky Added centos7/8 docker urls to the install test.
It seems official docker container for CentOS8 even cannot run yum update -y without error because of EOL.
I'll remove the test for centos8... On my CentOS8 installed from Disk worked. So this might be issue on docker image.
I've found a way for CentOS8 to run install-test-script without error.