scailium-installers

Bash installers and diagnostics for Scailium Engine (SQream DB), its HA / Pacemaker variant, and SQream Studio. Each script is standalone — copy the one you need onto the target server and run it there.

Script	Use it for
Scailium_Engine/scailium-install-v1.sh	Installing, upgrading and reconfiguring the engine — on-prem, AWS, OCI, MIG, metadata-only, CBO, Iceberg
Pacemaker/pcs_sqream_installer_V1.sh	The same engine in a Pacemaker HA cluster — master/slave nodes, node join/remove, worker add/delete
UI/sq-studio-install.sh	SQream Studio (Node.js + PM2), optionally behind an nginx HTTPS proxy
diag-sqream-v1.sh	Checking a host before or after an install (user, shell, sudoers, groups, limits, core dumps)

Requirements

• RHEL / CentOS 7, 8 or 9 (the scripts detect VERSION_ID and exit on anything else)
• root — run everything with sudo, except the Studio installer, which refuses to run as root
• An OS user named sqream — create it first, the installer exits if it is missing
• NVIDIA driver + nvidia-smi on GPU worker nodes
• The engine package as a <name>.<x>.<y>.tar.gz archive
• On RHEL 8/9 the -prepare step enables codeready-builder and highavailability repos via subscription-manager, so the host must be registered

Install the OS prerequisites (EPEL, monit, jq, kernel headers, build tools, chrony/ntp) with:

sudo ./scailium-install-v1.sh -prepare      # or: sudo ./pcs_sqream_installer_V1.sh -prepare

The engine must be stopped before installing or reconfiguring — the scripts abort if metadataserver.service is active.

Quick start

chmod +x scailium-install-v1.sh

sudo ./scailium-install-v1.sh -prepare                        # OS prerequisites
sudo ./scailium-install-v1.sh -f sqream-db-v2024.1.tar.gz     # fresh install
sudo ./scailium-install-v1.sh -S                              # review the summary

The installers are interactive — they prompt for the host IP, whether the node hosts the metadata server, the storage path, workers per GPU, and cudaMemQuota. Run them in a real terminal (tmux/screen for long sessions), not from a pipeline.

After installing, copy the logrotate rule for the engine logs:

sudo cp Scailium_Engine/scailium-services /etc/logrotate.d/

Scailium Engine flags

sudo ./scailium-install-v1.sh <flag> [package.tar.gz]

Flag	Description
-f <pkg>	Fresh install, monit-supervised
-u <pkg>	Upgrade the version and the storage; keeps existing configs
-a	Reconfigure an existing on-prem install
-aws <pkg> / -reaws	Install / reconfigure on AWS
-oci <pkg> / -reoci	Install / reconfigure on OCI
-mig <pkg> / -remig	Install / reconfigure with NVIDIA MIG support
-metadata <pkg> / -remetadata	Metadata server only, no workers
-cbo	Install the CBO components (cbo-backend, cbo-client, md-service)
-iceberg	Enable Iceberg support (requires CBO)
-E / -EA	Expert reconfiguration — edit worker JSON configs directly
-ldap	Configure an LDAP connection
-prepare	Install OS prerequisites
-L / -S	View the install log / the configuration summary
-h	Help

-p and -A also exist for Pacemaker installs, but use Pacemaker/pcs_sqream_installer_V1.sh for HA instead.

Pacemaker (HA) flags

sudo ./pcs_sqream_installer_V1.sh <flag> [package.tar.gz]

Flag	Description
-prepare	Prepare the OS for the engine and Pacemaker
-master <pkg> / -slave <pkg>	Install the first node / an additional node of the cluster
-remaster / -reslave	Reconfigure an existing master / slave node
-metadata <pkg>	Metadata server node for the cluster
-sqreamdb <pkg>	Engine only, on an existing cluster node
-join_gpu / -join_metadata	Join a new GPU / metadata node to the cluster
-add-workers / -delete-workers	Add or remove workers (run on the master)
-remove_node / -remove	Remove a node / a cluster resource
-status	Watch pcs status
-cbo	Install the CBO components
-S	View the configuration summary

Pacemaker creates a sqreamdb cluster with a lb_group resource group holding PublicVIP (IPaddr2), MS (metadata server) and LB (serverpicker). You are prompted for the hacluster credentials during setup.

The script's built-in -h text is out of date: it advertises -join, while the flag the script actually accepts is -join_gpu. It also omits -sqreamdb, -add-workers, -delete-workers, -remove_node, -status and -cbo.

SQream Studio flags

Run without sudo (the script calls sudo itself where needed):

Flag	Description
-pm2	Install Node.js, npm and PM2
-f <pkg>	Install Studio
-u <pkg>	Upgrade Studio
-nginx	Put an nginx HTTPS reverse proxy in front of Studio

Diagnostics

sudo ./diag-sqream-v1.sh

No flags — it walks the host checks in order (sqream user, shell, sudoers, groups, CUDA driver, security limits, core dumps), offers to configure the security limits and the core-dump path, and prints a summary of the errors it found at the end.

What gets installed where

Path	Contents
/usr/local/<package-version>	Extracted package, symlinked as /usr/local/sqream
/etc/sqream/	Worker configs (sqreamN_config.json), service env files, legacy config, license
/usr/lib/systemd/system/	sqreamN.service, metadataserver.service, serverpicker.service
/etc/monit.d/monitrc	monit process supervision for every worker and service
/var/log/sqream/	Runtime logs

Existing /etc/sqream is backed up to /etc/sqream_etc_backup_<timestamp> before a fresh install.

Ports: worker N listens on 4999+N (SSL 5099+N), metadata server on 3105, CBO client 6666, CBO backend 6665, md-service 6668.

Logs

File	Written by
/tmp/sqream-installV1.log	Engine installer (-L)
/tmp/sqreamdb-summary.log	Configuration summary (-S)
/tmp/pcs_sqream_installer_V1.log	Pacemaker installer
/tmp/sq-studio_install.log	Studio installer
/tmp/sqream-sqdiag.log	Diagnostics

Attach the relevant log when reporting an installation problem.