Skip to content

Latest commit

 

History

History
481 lines (364 loc) · 23.2 KB

File metadata and controls

481 lines (364 loc) · 23.2 KB

Manual steps required to setup machines

Adding firewall entries for Jenkins workers

Workers must be added to the firewall config before they will be able to connect to the Jenkins master.

You must be part of the infra group and have setup the ssh keys and config file beforehand.

To add an entry do the following:

  • ssh to the ci master: ssh ci
  • save the current config to a temporary file: iptables-save >foo
  • edit the temporary file with your favorite editor. Use one of the existing lines as a template and add a new entry at the end of the list of hosts just before the second COMMIT line near the end of the file.
  • restore the config from the temporary file: iptables-restore foo
  • remove the temporary file: rm foo

release-* machines

Once setup, they must have ~iojs/.ssh cloned from another machine, so they have the ssh setup and keys required to upload release artifacts to the nodejs.org web server. The result will be 2 files, an id_rsa containing a private key, and a config containing:

Host node-www
  HostName direct.nodejs.org
  User staging
  IdentityFile ~/.ssh/id_rsa

Both the config file and id_rsa should be owned and only readable by the user: chmod 700 .ssh && chmod 600 .ssh/*.

Its necessary to accept the known_hosts keys interactively on first ssh or the release builds will fail. After setting up .ssh, do something like this:

ssh node-www date
// ... accept the host keys

release-*container* machines

In the case of Docker container release hosts, the SSH configuration above works differently since the ~iojs home directories are elsewhere on the host machine. The Docker containers are started with /home/iojs inside the container mounted from /home/iojs/name-of-container/ on the host machine. Therefore, the above SSH configuration should take place in /home/iojs/name-of-container/.ssh/, with permissions set appropriately.

known_hosts can be primed and SSH tested from within the running containers:

  1. Find the running container ID using docker ps
  2. Enter the container using docker exec <containerid> -ti bash
  3. Run ssh node-www date (as above)

macOS release machines

Previous notes: #1393

Full Xcode

Xcode Command-line tools are not enough to perform a full notarization cycle, full Xcode must be installed manually.

As root:

  • Download Xcode: https://developer.apple.com/download/more/ - find non-beta version, open Developer Tools in browser, Networking tab, start download (then cancel), in Networking tab "Copy as cURL" (available in Chrome & FF)
  • Download onto release machine using the copied curl command (may need -o xcode.xip appended to curl command) to /tmp
  • Extract: xip --expand xcode.xip
  • Move Xcode.app directory to /Applications
  • xcode-select --switch /Applications/Xcode.app
  • xcodebuild -license - accept license

Signing certificates

  • Extract from secrets/build/release: dotgpg cat Apple\ Developer\ ID\ Node.js\ Foundation.p12.base64 | base64 -d > /tmp/Apple\ Developer\ ID\ Node.js\ Foundation.p12
  • Transfer to release machine (scp to /tmp)
  • sudo security import /tmp/Apple\ Developer\ ID\ Node.js\ Foundation.p12 -k /Library/Keychains/System.keychain -T /usr/bin/codesign -T /usr/bin/productsign -P 'XXXX' (where XXXX is found in secrets/build/release/apple.md) (security unlock-keychain -u /Library/Keychains/System.keychain may be required prior to running this command).

RHEL7-S390X

V8 build-tools

Copy /home/iojs/build-tools/{gn,ninja} from any other RHEL7-S390X host that has them to the newly provisioned one. Confirm gn --version is at least 1620.

If building from scratch, gn is built from https://gn.googlesource.com/gn, and ninja is built from https://github.com/ninja-build/ninja.

Note: If https://bugs.chromium.org/p/chromium/issues/detail?id=1029662 is resolved this will no longer be necessary.

devtoolset-6 install

First copy the rpms from a machine that already has them

scp -r test-ibm-rhel7-s390x-1:/data/devtoolset-6-s390x-rpms/ ~/devtoolset-6-s390x-rpms

Then copy them over to the target machine

scp -r ~/devtoolset-6-s390x-rpms {target host}:~/devtoolset-6-s390x-rpms/

Then install the rpms

yum install -y devtoolset-6-s390x-rpms/*

macOS

  1. Update Sudoers file:

this requires NOPASSWD to be added to the sudoers file to enable elevation

sudo visudo and change: %admin ALL = (ALL) ALL to %admin ALL = (ALL) NOPASSWD:ALL

  1. Allow ssh access
sudo systemsetup -setremotelogin on

AIX 7.2 Install

Most packages should be installed via ansible.

If there are any missing they should be installed via yum

What you do need to install manually is ccache

mkdir -p /opt/gcc-6.3 && cd /opt/gcc-6.3
curl -L https://ci.nodejs.org/downloads/aix/gcc-6.3-aix7.2.ppc.tar.gz | /opt/freeware/bin/tar -xzf -

ccache 3.7.4 on AIX 7.2

mkdir -p /opt/ccache-3.7.4 && cd /opt/ccache-3.7.4
curl -L https://ci.nodejs.org/downloads/aix/ccache-3.7.4.aix7.2.ppc.tar.gz | /opt/freeware/bin/tar -xzf -

Enable the AHA fs

For AIX 7 and 6.1, needed for the file watcher unit tests.

Add the following to /etc/filesystems:

/aha:
        dev             = /aha
        vfs             = ahafs
        mount           = true
        vol             = /aha

and then:

mkdir /aha
mount /aha

Install XL compilers

  1. Download 16.1.0 packages from: https://testcase.boulder.ibm.com (username: xlcomp4, password: ask @mhdawson)
  2. scp them to target:/opt/ibm-xlc
  3. on target:
cd /opt/ibm-xlc
uncompress 16.1.0.3-IBM-xlCcmp-AIX-FP003.tar.Z
uncompress IBM_XL_C_CPP_V16.1.0.0_AIX.tar.Z
installp -aXYgd ./usr/sys/inst.images -e /tmp/install.log all
inutoc
installp -aXgd ./ -e /tmp/install.log all
  1. Find compilers in /opt/IBM/xl[cC]/16.1.0/bin/

Fix "Missing" shared objects

On the 7.1 machines we were facing this issues when yum installed packages were not able to find some of the shared objects they needed

bash-5.0$ gmake -v
exec(): 0509-036 Cannot load program gmake because of the following errors:
        0509-022 Cannot load module /opt/freeware/lib/libintl.a(libintl.so.8).
        0509-150   Dependent module /usr/lib/libiconv.a(libiconv.so.2) could not be loaded.
        0509-152   Member libiconv.so.2 is not found in archive
        0509-022 Cannot load module make_64.
        0509-150   Dependent module /opt/freeware/lib/libintl.a(libintl.so.8) could not be loaded.
        0509-022 Cannot load module .

The fix is as follows:

sudo rm /usr/lib/libiconv.a && sudo ln -s /opt/freeware/lib/libiconv.a /usr/lib

Preparing gcc distributables

  1. download gcc-c++ (with dependencies) from bullfreeware.com

  2. scp 15412gcc-c++-6.3.0-1.aix7.2.ppc.rpm-with-deps.zip TARGET:/ramdisk0

    • Note: / is too small
  3. unzip 15412gcc-c++-6.3.0-1.aix7.2.ppc.rpm-with-deps.zip

  4. contained wrong libstdc++-9.1, so downloaded bundle for libstdc++ 6.3.0-1

  5. unpack the RPMs: $ for f in *gcc* *stdc*; do rpm2cpio $f | /opt/freeware/bin/cpio_64 -idmv; done

  6. Find absolute symlinks, and make them relative, example:

    $ find . -type l | xargs file
    ./opt/freeware/lib/gcc/powerpc-ibm-aix7.2.0.0/6.3.0/ppc64/libatomic.a: symbolic link to /opt/freeware/lib/gcc/powerpc-ibm-aix7.2.0.0/6.3.0/libatomic.a.
    ./opt/freeware/lib/gcc/powerpc-ibm-aix7.2.0.0/6.3.0/ppc64/libgcc_s.a: symbolic link to /opt/freeware/lib/gcc/powerpc-ibm-aix7.2.0.0/6.3.0/libgcc_s.a.
    ./opt/freeware/lib/gcc/powerpc-ibm-aix7.2.0.0/6.3.0/ppc64/libstdc++.a: symbolic link to /opt/freeware/lib/gcc/powerpc-ibm-aix7.2.0.0/6.3.0/libstdc++.a.
    ./opt/freeware/lib/gcc/powerpc-ibm-aix7.2.0.0/6.3.0/ppc64/libsupc++.a: symbolic link to /opt/freeware/lib/gcc/powerpc-ibm-aix7.2.0.0/6.3.0/libsupc++.a.
    ./opt/freeware/lib/gcc/powerpc-ibm-aix7.2.0.0/6.3.0/pthread/ppc64/libatomic.a: symbolic link to /opt/freeware/lib/gcc/powerpc-ibm-aix7.2.0.0/6.3.0/pthread/libatomic.a.
    ./opt/freeware/lib/gcc/powerpc-ibm-aix7.2.0.0/6.3.0/pthread/ppc64/libgcc_s.a: symbolic link to /opt/freeware/lib/gcc/powerpc-ibm-aix7.2.0.0/6.3.0/pthread/libgcc_s.a.
    ./opt/freeware/lib/gcc/powerpc-ibm-aix7.2.0.0/6.3.0/pthread/ppc64/libstdc++.a: symbolic link to /opt/freeware/lib/gcc/powerpc-ibm-aix7.2.0.0/6.3.0/pthread/libstdc++.a.
    ./opt/freeware/lib/gcc/powerpc-ibm-aix7.2.0.0/6.3.0/pthread/ppc64/libsupc++.a: symbolic link to /opt/freeware/lib/gcc/powerpc-ibm-aix7.2.0.0/6.3.0/pthread/libsupc++.a.
    
    bash-5.0# pwd
    /ramdisk0/aixtoolbox/opt/freeware/lib/gcc/powerpc-ibm-aix7.2.0.0/6.3.0/ppc64
    
    bash-5.0# ln -fs ../libatomic.a ../libgcc_s.a ../libstdc++.a ../libsupc++.a ./
    
    bash-5.0# find . -type l | xargs file
    ./ppc64/libatomic.a: archive (big format)
    ./ppc64/libgcc_s.a: archive (big format)
    ./ppc64/libstdc++.a: archive (big format)
    ./ppc64/libsupc++.a: archive (big format)
    ./pthread/ppc64/libatomic.a: symbolic link to /opt/freeware/lib/gcc/powerpc-ibm-aix7.2.0.0/6.3.0/pthread/libatomic.a.
    ./pthread/ppc64/libgcc_s.a: symbolic link to /opt/freeware/lib/gcc/powerpc-ibm-aix7.2.0.0/6.3.0/pthread/libgcc_s.a.
    ./pthread/ppc64/libstdc++.a: symbolic link to /opt/freeware/lib/gcc/powerpc-ibm-aix7.2.0.0/6.3.0/pthread/libstdc++.a.
    ./pthread/ppc64/libsupc++.a: symbolic link to /opt/freeware/lib/gcc/powerpc-ibm-aix7.2.0.0/6.3.0/pthread/libsupc++.a.
    
    bash-5.0# cd pthread/ppc64/
    
    bash-5.0# ln -fs ../libatomic.a ../libgcc_s.a ../libstdc++.a ../libsupc++.a ./
    
    bash-5.0# file *.a
    libatomic.a: archive (big format)
    libgcc.a: archive (big format)
    libgcc_eh.a: archive (big format)
    libgcc_s.a: archive (big format)
    libgcov.a: archive (big format)
    libstdc++.a: archive (big format)
    libsupc++.a: archive (big format)
    
  7. Move to target location and create a tarball with no assumptions on leading path prefix:

    $ mkdir /opt/gcc-6.3
    $ cd /opt/gcc-6.3
    $ mv .../opt/freeware/* ./
    $ tar -cvf ../gcc-6.3-aix7.2.ppc.tar *
    

Example above was for 6.3.0, but process for 4.8.5 is identical, other than the version numbers.

Example search for 4.8.5 gcc on bullfreeware:

Preparing ccache distributables

Notes:

  • AIX tar doesn't know about the "z" switch, so use GNU tar.

  • Build tools create 32-bit binaries by default, so explicitly create 64-bit ones.

    $ curl -L -O https://github.com/ccache/ccache/releases/download/v3.7.4/ccache-3.7.4.tar.gz
      % Total    % Received % Xferd  Average Speed   Time    Time     Time  Current
    				 Dload  Upload   Total   Spent    Left  Speed
    100   607    0   607    0     0   3281      0 --:--:-- --:--:-- --:--:--  3281
    100  490k  100  490k    0     0   586k      0 --:--:-- --:--:-- --:--:-- 60.4M
    $ /opt/freeware/bin/tar -xzf ccache-3.7.4.tar.gz
    $ cd ccache-3.7.4
    $ ./configure CC="gcc -maix64" && gmake
    $ mkdir -p /opt/ccache-3.7.4/libexec /opt/ccache-3.7.4/bin
    $ cp ccache /opt/ccache-3.7.4/bin
    $ cd /opt/ccache-3.7.4/libexec
    $ ln -s ../bin/ccache c++
    $ ln -s ../bin/ccache cpp
    $ ln -s ../bin/ccache g++
    $ ln -s ../bin/ccache gcc
    $ ln -s ../bin/ccache gcov
    $ cd cd /opt/ccache-3.7.4
    $ tar -cf /opt/ccache-3.7.4.aix7.2.ppc.tar.gz *
    

Windows (Azure/Rackspace)

In order to get Windows machines to a state where Ansible can be run against them, some manual steps need to be taken so that Ansible can connect.

Machines should have:

  • Remote Desktop (RDP) enabled, the port should be listed with the access credentials if it is not the default (3389).
  • PowerShell access enabled, the port should be listed with the access credentials if it is not the default (5986).

Control machine (where Ansible is run)

Install the pywinrm pip module: pip install pywinrm

Target machines

The preparation script needs to be run in PowerShell (run as Administrator):

iwr -useb https://raw.githubusercontent.com/ansible/ansible/devel/examples/scripts/ConfigureRemotingForAnsible.ps1 | iex

Test the connection to the target machine with ansible HOST -m win_ping -vvvv. If there is any issue, please refer to the official Ansible documentation in Setting up a Windows Host.

jenkins-workspace

The hosts labelled jenkins-workspace are used to "execute" the coordination of Jenkins jobs. Jenkins uses them to do the initial Git work to figure out what needs to be done before farming off to the actual test machines. These machines are lower powered but have large disks so they can waste space with the numerous Git repositories Jenkins will create in this process. The use of these hosts takes a load off the Jenkins master and prevents the Jenkins master from filling up its disk with Git repositories.

Note that not all jobs can use jenkins-workspace servers for execution, some are tied to other hosts.

The jenkins-workspace hosts are setup as standard Node.js nodes but are only given the jenkins-workspace label. After setup, they require the following manual steps:

  • Download the Coverity Build Tool for Linux x64 at https://scan.coverity.com/download (requires a Coverity login)
  • Extract to /var, e.g. so the resulting directory looks like /var/cov-analysis-linux64-2017.07/ or similar
  • Ensure that the node-coverity-daily job matches the path used in its explicit PATH setting

Docker hosts

The hosts that run Docker images for "sharedlibs", Alpine Linux and a few other dedicated systems (hosts identified by grep _docker-x64- inventory.yml) don't have Docker image reload logic built in to Ansible. Changes to Docker images (adding, deleting, modifying) involve some manual preparation.

The general steps are:

  1. Stop the concerned Jenkins systemd service(s) (sudo systemctl stop jenkins-test-$INSTANCE)
  2. Disable the concerned Jenkins systemd service(s) (sudo systemctl disable jenkins-test-$INSTANCE)
  3. Remove the Jenkins systemd service configuration (rm /lib/systemd/system/jenkins-test-$INSTANCE.service)
  4. systemctl daemon-reload to reload systemd configuration from disk
  5. systemctl reset-failed to remove the disabled and removed systemd service(s)
  6. Clean up unnecessary Docker images (docker system prune -fa to clean everything up, or just docker rm for the images that are no longer needed and a lighter docker system prune after that to clean non-tagged images).

Steps 3-5 may not be strictly necessary in the case of a simple modification as the existing configurations will be reused or rewritten by Ansible anyway.

To completely clean the Jenkins and Docker setup on a Docker host to start from scratch, either re-image the server or run the follwing commands:

systemctl list-units -t service --plain --all jenkins* | grep jenkins-test | awk '{print $1}' | xargs -l sudo systemctl stop
systemctl list-units -t service --plain --all jenkins* | grep jenkins-test | awk '{print $1}' | xargs -l sudo systemctl disable
sudo rm /lib/systemd/system/jenkins-test-*
sudo systemctl daemon-reload
sudo systemctl reset-failed
sudo docker system prune -fa

To do this across multiple hosts, it can be executed with parallel-ssh like so:

parallel-ssh -i -h /tmp/docker-hosts 'systemctl list-units -t service --plain --all jenkins* | grep jenkins-test | awk '\''{print $1}'\'' | xargs -l sudo systemctl stop'
parallel-ssh -i -h /tmp/docker-hosts 'systemctl list-units -t service --plain --all jenkins* | grep jenkins-test | awk '\''{print $1}'\'' | xargs -l sudo systemctl disable'
parallel-ssh -i -h /tmp/docker-hosts 'sudo rm /lib/systemd/system/jenkins-test-*'
parallel-ssh -i -h /tmp/docker-hosts 'sudo systemctl daemon-reload'
parallel-ssh -i -h /tmp/docker-hosts 'sudo systemctl reset-failed'
parallel-ssh -i -h /tmp/docker-hosts 'sudo docker system prune -fa'

Note that while this is being done across all Docker hosts, you should disable node-test-commit-linux-containered to avoid a queue and delays of jobs. The Alpine Linux hosts under node-test-commit-linux will also be impacted and may need to be manually cancelled if there is considerable delay. Leaving one or more Docker hosts active while reloading others will alleviate the need to do this.

SmartOS

Joyent SmartOS machines use libsmartsshd.so for PAM SSH authentication in order to look up SSH keys allowed to access machines. Part of our Ansible setup removes this so we can only rely on traditional SSH authentication. Therefore, it is critcal to put nodejs_test_* public keys into $USER/.ssh/authorized_keys as appropriate or access will be lost and not recoverable after reboot or sshd restart (part of Ansible setup).

Raspberry Pi

Raspberry Pi configuration is integrated into the standard Ansible playbooks and will run properly when the right hosts are executed.

The current configuration relies upon an NFS boot and NFS root architecture, although it should still be possible to connect a non-NFS Raspberry Pi to the Node.js CI and the Ansible playbooks are intended to be friendly to non-NFS hosts. It's possible that current Ansible scripts don't properly account for non-NFS hosts since these are not regularly included so some adjustments may be necessary.

This document covers much of the process we use: https://www.raspberrypi.org/documentation/hardware/raspberrypi/bootmodes/net_tutorial.md

NFS boot

  • The SD card in the Raspberry Pi should have an up to date bootcode.bin (found in the FAT partition of the Raspbian images and comes via regular updates) and ideally an updated firmware. It is possible to boot newer Pi's without SD card but the SD card is necessary for older models and our Ansible setup uses it for local swap space.
  • Upon boot, the bootcode will cause the Pi to reach out for an initial DHCP discovery. A DHCP server should be configured to respond appropriately for that device's address along with some NFS boot signals, such as this configuration for dnsmasq:
enable-tftp
tftp-root=/var/tftpd
pxe-service=0,"Raspberry Pi Boot"
  • The tftp server on the same DHCP server should be able to respond to boot requests.
  • After looking in the root of the tftp server, the Pi will attempt to load files from a subdirectory named by its serial number. Obtain the serial number from /proc/cpuinfo from a running Pi, take the last 8 characters from the Serial field and use that to store individual boot files.
  • Copy the entirety of the boot partition of the Raspbian disk image into a subdirectory for each Pi. Symlinks are acceptable to map serial numbers to the names of the Pi's to keep them organised properly.
  • Edit cmdline.txt inside the tftp subdirectory for each file and replace the contents with:

`` modprobe.blacklist=bcm2835_v4l2 root=/dev/nfs nfsroot=NFS_ROOT_SERVER_IP:/NFS_ROOT_FOR_THIS_PI,vers=3 rw ip=dhcp rootwait elevator=deadline


* Replacing `NFS_ROOT_SERVER_IP` and `NFS_ROOT_FOR_THIS_PI` as appropraite. It is assumed that NFSv3 is used for the root server. The `modprobe.blacklist` is for a sound driver that has been causing problems in most 2019 versions of the Raspbian kernels (Stretch and Buster), this may be fixed at a later day and be unnecessary. Booting would freeze late in the process without this driver removed.
* The DHCP / tftp server should export each of the boot directories via NFS so they can be mounted by the Pi's as /boot/ which will allow the files to be updated during system updates.

### NFS root

An NFS root server can be separate from NFS boot server, and could be different for each Pi.

* The NFS root server should export a shared .ccache directory to be mounted by all Pi's, so it should be exported in such a way as to be permissive with IP addresses. The export should have roughly these options: `(async,rw,all_squash,anonuid=1001,anongid=1002,no_subtree_check)`.
* The NFS root server should export a root directory for each Pi. The IP of the server along with the path to the directory should be put in cmdline.txt on the NFS boot server. The exports should have roughly these options: `(rw,sync,no_subtree_check,no_root_squash)`/
* The ext4 partition of the Raspbian image file (second partition, not the boot FAT partition) should be extracted into this root directory.
* `etc/fstab` in the root directory should be edited to make it NFS compatible. Remove the existing `/boot` and `/` entries and replace them with:

/dev/mmcblk0p1 /mnt/mmcblk0p1 vfat defaults 0 0
NFS_ROOT_SERVER_IP:PATH_TO_SHARED_CCACHE_DIRECTORY /home/iojs/.ccache nfs4 rw,exec,async,noauto 0 0
NFS_BOOT_SERVER_IP:PATH_TO_TFTP_BOOT_EXPORT /boot nfs4 nfsvers=3,rw,noexec,async,noauto 0 0
  • Ansible should also perform checks on /etc/fstab so these modifications may not be strictly necessary but it is helpful to have first-boot be into an appropriate state.
  • Mounting / is done during the NFS boot process so is omitted from /etc/fstab.
  • The SD card is mounted at /mnt/mmcblk0p1 and is assumed to be in this location by the Ansible scripts for swap file creation.
  • /boot is this Pi's tftp boot directory from the NFS boot server.
  • When powered on, the Pi should perform all mount steps and present with the standard initial Raspbian boot & login. Note that SSH is not enabled by default and this needs to be done manually before you can remotely access it. To streamline setup, these additional steps can be performed on an initial Pi and then its root directory copied to all other root directories with only minor modifications to /etc/fstab required:
    • Enable SSH with raspi-config
    • Add the nodejs_build_test public SSH key to ~pi/.ssh/authorized_keys (with appropriate permissions).
    • Change the default password for user pi to remove insecurity. This could even be disabled entirely since the SSH key is in place.

After these steps are performed and the Pi's are running, Ansible can be run to finish setup. A reboot is recommended after initial setup to ensure the environment is configured correctly (locale and other settings that are changed).