SafeScale v22.11.9

New Features

• includes a new azure driver based on terraform 1.4.2. (usable but not ready for production)
• includes a new framework to help create new providers based on terraform instead of API calls.

Bugfixes.

• fixes incompatibility issues detected operating clusters created with previous SafeScale versions.
• fixes some metadata corruption issues
• cleans documentation
• fixes several problems in input validation
• bumps dependencies
• removed dead code
• removed obsolete drivers and features

Full Changelog: v22.11.6...v22.11.9

SafeScale v22.11.6

Bugfix Release.

• fixes incompatibility issues detected operating clusters created with previous safescale versions.
• images based on Ubuntu 22.04 are now supported
• fixes several disk sizing problems creating clusters
• fixes several problems in machine provisioning scripts
• introduces grpc-web interface
• fixes some metadata corruption issues
• removes python3-setuptools dependency in provisioning scripts
• improves command 'safescale tenant list'
• cleans documentation
• changes docker build scripts
• fixes problems validating inputs
• bumps dependencies

What's Changed

• Release v22.11.6 by @oscarpicas in #626

Full Changelog: v22.11.5...v22.11.6

SafeScale v22.11.5

Stabilization release

• SafeScale can now reliably create clusters with large numbers of nodes (previously node failures halted the cluster creation, now SafeScale keeps building until the node requirements are met)

Also:

• label and tags now use the label/tag/metadata functions offered by each provider, so they are visible in the web console of each provider and can be used by K8S and other tooling
• fixed problems with security groups on some providers
• fixed problems installing docker on some combinations of provider/OS
• now you can choose the number of masters in your cluster with the "count" keyword
• fixes several metadata inconsistencies
• ansible inventory was not updated correctly when nodes were removed
• fixed problems tagging machines, now all tags are displayed with a "safescale host inspect"
• fixed problems in operations due to improper cache invalidation
• fixes wrong and outdated documentation
• removed unimplemented commands that displayed the message "not yet implemented" to users
• fixed improper start/stop of machines due to checking with metadata and not looking at the status exposed by the provider.
• fixed problem with "safescale host ls" displaying non-existing machines
• fixed deadlocks and panics

Known limitations

• Mount of Google Object Storage Bucket (still...) not yet implemented

What's Changed

• Bugfixes by @oscarpicas in #557
• Security Group Bugfixes by @oscarpicas in #565
• Copyright update by @oscarpicas in #570
• Cleanup by @oscarpicas in #572
• Bugfixes by @oscarpicas in #581
• Bugfixes by @oscarpicas in #583
• Bugfix Cache by @oscarpicas in #585
• Cleanup by @oscarpicas in #586
• Feature changes by @oscarpicas in #591
• Cleanup, Minor bugfixes, Documentation and Integration Tests by @oscarpicas in #592
• Bugfixes by @oscarpicas in #596
• Bugfixes by @oscarpicas in #597
• Feature #91 by @oscarpicas in #598
• Bugfix by @oscarpicas in #600
• Ansible for cluster feature by @oscarpicas in #601
• Bugfix: Metadata by @oscarpicas in #602
• Removed dead code by @oscarpicas in #605
• Bugfix: docker by @oscarpicas in #606
• Bugfixes: cluster by @oscarpicas in #607
• Several Bugfixes by @oscarpicas in #609
• Bugfixes and Doc updates by @oscarpicas in #611

Full Changelog: v22.11.1...v22.11.5

Safescale v22.11.1

Changes since v22.11.0

bugfixes

• gateway name is ignored: FIXED
• various metadata inconsistencies: FIXED

SafeScale v22.11.0

Stabilization release

• Cluster creation with lot of nodes are working a lot better (tested with ~100 nodes)

Bugfixes

• Fixed some specific OVH issues

Known limitations

• Mount of Google Object Storage Bucket (still...) not yet implemented

Known bugs

• documentation has been improved but still needs work

SafeScale v22.06.1

Changes since v22.06.0

bugfixes

• panic occurring in provider FlexibleEngine (stack huaweicloud) when provider uses too much time to create hosts, leading to a nil dereference: FIXED

SafeScale v22.06.0

New features

• Added resources Label and Tag

  A Label is a way to categorize Hosts. Each created Label has a default value, and can be bound to many Hosts. During the bind, the Host can override the Label default value.
  A Tag is a Label without a value. Each created Tag can be bound to many Hosts.

  Currently, this is only a SafeScale metadata magic. In a future release, Label and Tags will at least have their Cloud Provider counterparts updated. Consideration is being given to extend the synchronisation of Labels/Tags to other components (groups in Ansible inventory, Kubernetes annotations, ...).

Developer interest

• New integration test framework with ability to execute specific test and to avoid human mistakes due to duplication of code provider by provider (now the same code is used everywhere).

Bugfixes

• Fixed some specific Outscale issues

Known limitations

• Mount of Google Object Storage Bucket (still...) not yet implemented

Known bugs

• documentation has been improved but still needs work

SafeScale v21.11.1

!!! Warning !!!

Format of SafeScale metadata has evolved. It's strongly advised to not use the binaries of this release with existing SafeScale resources. However, if you are adventurous, there is a metadata upgrade procedure proposed when using safescale tenant set <tenant>, that still need more testings (there are some situations where the migration may fail, leaving metadata in inconsistent state). Use at your own risk !!! It's strongly recommended to realize a metadata backup; rclone may be of good use for that.
Next revision should remove this warning with further tested migration code (hopefully v21.11.1).

What's new since v21.11.0

This release contains mainly bugfixes:

• #90 Cluster subnet CIDR must not overlap with Docker default subnet: FIXED
• #114 safescale cluster expand command blocks after image selection by safescaled: FIXED
• #115 Unable to create SSD Volume on FlexibleEngine: FIXED
• #149 sizing parameters ignored upon cluster creation: FIXED
• #173 Disable Flag gateway-failover not working correctly: FIXED
• #195 safescale --server flag not honoured: FIXED
• CLI flags are ignored if placed after arguments (since upgrade to urfave/cli v2): FIXED
• OVH: templates i1.* contain far more nvme disk than reported by the API: FIXED
• system disk sizing not always honored for all Cloud Providers: FIXED
• in cluster creation, using component count in --master-sizing should trigger an error: FIXED
• file descriptor leakage due to unclosed http.Responses/Connections: FIXED
• wrong defaults for timings: FIXED
• cluster cannot be inspected while building: FIXED
• keepalived configuration invalid on secondary gateway, disabling default route not working if primary gateway stops: FIXED
• "ssh connect" does not even try to use the secondary gateway if primayr fails: FIXED
• multiple timings degradations: FIXED
• undeletable cluster: FIXED
• huge docker install time: FIXED
• some panics remains: FIXED

• various fixes in bash scripts

Developer interest

• Reverted urface/cli from v2 to v1 (issue with flags after arguments in v2)

Known limitations:

• Mount of Google Object Storage Bucket not yet implemented

Known bugs

• documentation has been improved but still needs work

SafeScale v21.11.0

!!! Warning !!!

Format of SafeScale metadata has evolved. It's strongly advised to not use the binaries of this release with existing SafeScale resources. However, if you are adventurous, there is a metadata upgrade procedure proposed when using safescale tenant set <tenant>, that still need more testings (there are some situations where the migration may fail, leaving metadata in inconsistent state). Use at your own risk !!! It's strongly recommended to realize a metadata backup; rclone may be of good use for that.
Next revision should remove this warning with further tested migration code (hopefully v21.11.1).

What's new

• Introducing Security Group support. Please refer to documentation for more details (Security Group resource, usage).
• SafeScale Network abstraction evolves: where in previous releases, a SafeScale Network encapsulated a resource couple (network/VPC + subnet), now SafeScale abstracts both Network and Subnet. Previous behaviour is kept as default (creates a Network AND a Subnet with the same names), with the ability to create an empty SafeScale Network in which multiple Subnets can be created
• Significant improvements in tolerance to temporary communication failures (loss of connection, failed DNS resolution, variable latency, etc.); when possible, retries are attempted instead of immediate failure.
• Merged scanner binary into safescale/safescaled. Use is now safescale tenant scan <tenant name>

Security Fixes

• in previous releases of SafeScale, private SSH key may be readable from some Cloud Provider metadata services, from Host itself only. Now, even if a private SSH key is still readable, this key is used only once for first connection to the Host and then immediately replaced by a new private SSH key.

Developer interest

• Refactoring done to separate IaaS code from SafeScale metadata (previously intimately mixed; this consumed a lot of time actually)
• Moved SafeScale Platform code (Cluster, Features) from safescale to safescaled
• Added metadata versioning (idea being to be able to migrate metadata of previous releases when necessary; ie warning message below)
• Replaced Jaro-Winkler image selection algorithm by WagnerFischer (Levenshtein Distance kind of algorithm) with pre- and post-processing
• Replaced standard json package use with github.com/json-iterator/go
• Refactored Object Storage Bucket mount around the use of rclone instead of s3fs

Known limitations:

• Mount of Google Object Storage Bucket not yet implemented

Known bugs

• CLI flags are ignored if placed after parameters (since upgrade to urfave/cli v2), ie safescale network create --cidr 192.168.1.0/24 my-network interprets --cidr correctly, while safescale network create my-network --cidr 102.168.1.0/24 will not
• OVH: templates i1.* contain far more nvme disk than reported by the API, costing a lot more than expected if selected
• documentation has been improved but still needs work
• system disk sizing not always honored for all Cloud Providers

SafeScale v20.06.2

Fixed bugs

• safescale cluster shrink leaves metadata of deleted nodes, leading to errors on next safescale cluster commands; FIXED
• panic may occur in marginal cases when parsing JSON; FIXED
• default route not configured correctly on gateway in marginal cases; FIXED
• safescale ssh connect -u <user>does not ask for password; FIXED
• weaknesses in error handling of some package installations in deployment scripts; FIXED