@usrbinkat.io

@usrbinkat.io

4427 Followers 5374 Following

🧙‍♀️ Open Sourceress 🚀 Ms. (f)Rizzle @Cisco 🏳️‍⚧️ Blame for Oops & Ops my own 🤗 Seize the means of computing 👩‍💻 K8s/AWS/Azure Platform Engineer 📖 https://blog.usrbinkat.io/en/page/about 🍿Watching Dystopia IRL 🧠 Neuro Spicy Autist 📍 Sacramento ☀

A screenshot shows terminal output of a Ceph Storage Benchmark running on the NVMe tier from node op1 in a tmux session. The benchmark configuration header shows pod bench-nvme running tests for seqwrite, seqread, randwrite, randread, and mixed workloads with 25G file size, IO depth 32, 4 jobs, 30s runtime with time-based tests, using O_DIRECT, mounted at /data. The benchmark date is 2026-06-25. Five test results display in sequence. Sequential Write achieves 963.09 MB/s at 963 IOPS with 132.882ms latency. Sequential Read reaches 1796.80 MB/s at 1797 IOPS with 71.199ms latency. Random 4K Write delivers 46.13 MB/s at 11812 IOPS with 10.834ms latency. Random 4K Read hits 191.25 MB/s at 48961 IOPS with 2.612ms latency. Mixed 70/30 4K workload shows 34.43 MB/s reads at 8816 IOPS with 3.919ms latency and 34.43 MB/s writes at 8816 IOPS with 10.595ms latency. Below the results, a formatted markdown Ceph Storage Benchmark Report begins with parameter and results tables. The benchmark validates the replacement WD_BLACK SN850X drives are performing well in the rebuilt Ceph cluster, with nearly 1.8 GB/s sequential reads and almost 49K random read IOPS through the full Ceph stack including network replication. These numbers reflect real distributed storage performance rather than raw drive speeds, with the sequential write throughput around 963 MB/s indicating the overhead of Ceph's replication factor of 2 across the network while random 4K performance demonstrates the SN850X's strong small-block capability translating effectively through the Rook-Ceph CSI layer on Talos Linux.
A screenshot shows K9s terminal UI displaying 16 Kubernetes StorageClasses on the optiplexprime-konduktor-admin cluster running Kubernetes 1.36.2. A keyboard shortcut menu is visible in the header showing options for Delete, Describe, Edit, Help, Jump Owner, and YAML. The storageclass view lists resources in a table with columns for Name, Provisioner, ReclaimPolicy, and VolumeBindingMode. The storage classes span four Ceph storage backends. S3 bucket classes include ceph-bucket-s3-archive highlighted in blue, ceph-bucket-s3-nvme, and ceph-bucket-s3-ssd, all using rook-ceph.ceph.rook.io/bucket provisioner. NVMeoF classes include ceph-nvmeof-hdd, ceph-nvmeof-nvme, and ceph-nvmeof-ssd using rook-ceph.nvmeof.csi.ceph.com with Immediate binding. RBD block classes appear in pairs with Delete and Retain reclaim policies for hdd, nvme, and ssd tiers using rook-ceph.rbd.csi.ceph.com, with rook-ceph-block-nvme marked as cluster default. CephFS classes include rook-cephfs-hdd, rook-cephfs-nvme, and rook-cephfs-ssd using rook-ceph.cephfs.csi.ceph.com. A hostpath-provisioner for KubeVirt also appears. All non-NVMeoF classes use WaitForFirstConsumer binding. This comprehensive storage class matrix demonstrates the full tiered storage architecture now operational after the NVMe drive replacement and Pulumi deployment. The NVMe tier set as cluster default reflects prioritizing the new WD_BLACK SN850X drives for general workloads, while the systematic coverage of block, filesystem, object, and NVMeoF access modes across HDD, SSD, and NVMe device classes provides Kubernetes workloads with granular performance selection entirely managed through declarative IaC.
A screenshot shows K9s terminal UI displaying Ceph block pools in the rook-ceph namespace on Kubernetes 1.36.2. The K9s header shows context usrbinkat-optiplexprime-konduktor-admin in RW mode with K9s version 0.51.0, and a namespace selector listing all, authelia-system, default, rook-ceph, and braincraft. The ASCII K9s dog logo appears in the upper right corner. The cephblockpools(all)[4] view lists four block pools in a table with columns for Namespace, Name, Phase, Type, Failuredomain, and Age. All four pools show rook-ceph namespace, Ready phase, Replicated type, and host failure domain. The pools are hdd-blockpool at 28h, nvme-blockpool highlighted in blue at 2m59s indicating it was just created, nvmeof at 26h, and ssd-blockpool at 28h. The bottom bar shows navigation breadcrumbs for customresourcedefinition and cephblockpool. The freshly created nvme-blockpool at just under 3 minutes old confirms the Pulumi deployment successfully provisioned the new NVMe-tier block storage on the replacement WD_BLACK drives. The four-tier block pool architecture spanning HDD, SSD, NVMe, and NVMeoF device classes demonstrates granular storage tiering within a single Ceph cluster, enabling workload-appropriate performance selection through Kubernetes StorageClasses all managed declaratively through Pulumi IaC.
A screenshot shows Pulumi executing a stack update for optiplex-admin-storage in a tmux terminal session. The command pulumi up --stack optiplex-admin-storage --skip-preview=true --refresh=true runs from usrbinkat@mithril connected to the optiplexprime-konduktor-admin:default braincraft main context. A browser link points to app.pulumi.com/usrbinkat/infrastructure/optiplex-admin-storage/updates/26. The output displays a hierarchical resource tree with columns for Type, Name, Status, and Info. The tree shows the pulumi:pulumi:Stack at the root refreshing with a resource warning, branching into dozens of Kubernetes resources. Visible components include k8s-provider, zot-registry chart with its ServiceAccount, ConfigMap, Service, and Secret, snapshot-controller with RBAC and deployment resources, cloudnative-pg-operator ConfigMap, redis-operator chart with deployment and RBAC bindings, and an extensive rook-ceph-operator-chart section dominating the output with dozens of ServiceAccounts, Roles, ClusterRoles, ClusterRoleBindings, CustomResourceDefinitions, and RoleBindings for ceph-csi, rook-ceph-osd, rook-ceph-mgr, rook-ceph-system, objectstorage-provisioner, and other Ceph components. Status column shows refresh timings ranging from 0s to 1s. A kubeconfig-store dynamic resource appears at the bottom. This captures the Pulumi IaC deployment applying the updated storage configuration with the new WD_BLACK NVMe device definitions and NVMe-tier CephFS and S3 configurations. The extensive Rook Ceph operator resource tree reflects the complexity of declaratively managing distributed storage through pure infrastructure as code, with every Kubernetes resource from RBAC policies to CRDs version-controlled and deployed programmatically without a single helm or kubectl command.
A screenshot shows Claude Code in a tmux terminal session displaying a YAML diff adding an NVMe-tier RADOS Gateway S3 object storage configuration to Pulumi.optiplex-admin-storage.yaml. Green highlighted lines from 546 to 569 define the new block commented as NVMe tier - Hot data with 1TB WD SN850X TLC. The configuration specifies enabled true, name s3-nvme, device_class nvme, instances 1, port 80, metadata_pool_replicated_size 3, data_pool_replicated_size 2, data_pool_erasure_coded false, and resource limits of 250m/1 CPU and 512Mi/2Gi memory. The route section enables an httproute with hostname s3.nvme.optiplexprime.home.arpa, additional hostname s3.nvme.braincraft.io, gateway_type envoy, gateway_name envoy-gateway in envoy-gateway-system namespace, and gateway_listener_name https. Below, the existing SSD tier S3 configuration with s3-ssd remains visible. Claude Code prompts with option 1 Yes selected. This adds a dedicated NVMe-backed S3-compatible object storage endpoint alongside the existing SSD tier, creating a hot-data storage class on the replacement WD_BLACK drives. The Envoy gateway routing with both local .home.arpa and public braincraft.io hostnames enables S3 API access from both the homelab network and external services. The dual-tier S3 architecture separates hot and warm data across NVMe and SSD device classes, with metadata at triple replication for durability and data at double replication to balance redundancy with the 1TB per-node NVMe capacity.
A screenshot shows Claude Code in a tmux terminal session displaying a YAML diff adding a new NVMe-tier CephFS configuration to Pulumi.optiplex-admin-storage.yaml. Green highlighted lines from 475 to 490 show the new block being added with comment NVMe tier CephFS 1TB WD SN850X TLC. The configuration defines enabled true, name rook-cephfs-nvme, filesystem name cephfs-nvme, pool name data0, device class nvme, is_default false, replicated_size 2, metadata_replicated_size 3, reclaim_policy Delete, allow_volume_expansion true, volume_binding_mode WaitForFirstConsumer, mds_active_count 1, mds_standby_count 1, and pg_autoscale_mode_data and metadata both set to on. Below the new block, the existing SSD tier CephFS configuration beginning with rook-cephfs-ssd remains visible for context. Claude Code prompts with option 1 Yes selected. This adds a dedicated NVMe-backed CephFS filesystem alongside the existing SSD tier, creating a high-performance storage class on the newly installed WD_BLACK SN850X drives. CephFS provides ReadWriteMany access mode essential for KubeVirt live migration, where virtual machine disks must be simultaneously accessible from both source and destination nodes during migration. The replicated_size of 2 balances redundancy with capacity across the three-node cluster, while WaitForFirstConsumer binding ensures volumes are provisioned on nodes with available NVMe-class OSDs.
A screenshot shows Claude Code in a tmux terminal session displaying a YAML diff for the file Pulumi.optiplex-admin-storage.yaml. The diff shows changes across three Ceph storage nodes: op1 at 10.0.0.33, op2 at 10.0.0.34, and op3 at 10.0.0.35. For each node, red highlighted lines show commented-out old Sabrent NVMe devices marked as offline for maintenance, while green highlighted lines show their replacements: new WD_BLACK SN850X 1TB TLC drives with full /dev/disk/by-id paths. Each node retains an unchanged secondary SSD device alongside the replaced NVMe. The device paths show the naming convention shift from nvme-Sabrent identifiers to nvme-WD_BLACK_SN850X_HS_1000GB with unique serial numbers for each of the three drives. Node op1 includes an additional comment noting the 2TB Sabrent was replaced and is now the Talos system install disk. At the bottom, Claude Code prompts Do you want to make this edit to Pulumi.optiplex-admin-storage.yaml? with option 2 selected: Yes, allow all edits during this session. This captures the infrastructure-as-code approach to Ceph cluster recovery using Pulumi Python IaC, where replacing failed drives means updating YAML configuration with new device identifiers rather than running manual kubectl or ceph commands. Claude Code is assisting with the declarative storage infrastructure update, embodying the no-helm-no-kubectl philosophy where all cluster state is managed through version-controlled Pulumi definitions deployed programmatically.
A screenshot shows the Sidero Omni web interface displaying the Disks tab for node op1 in a dark-themed dashboard. The interface shows two NVMe drives. The first, nvme0n1 at /dev/nvme0n1, is a 1.0 TB nvme drive showing entirely as Unallocated 931.51 GB in gray, representing the newly installed replacement WD_BLACK SN850X ready for Ceph OSD provisioning. The second drive, nvme1n1 at /dev/nvme1n1, is a 2.0 TB nvme drive with a horizontal bar visualization showing partition layout. Four partitions are listed below: EFI at /dev/nvme1n1p1 with vfat filesystem, unknown encryption, and 2.2 GB size; META at /dev/nvme1n1p2 with talosmeta filesystem, encryption disabled, at 1.0 MB; STATE at /dev/nvme1n1p3 with xfs filesystem, encryption disabled, with UUID shown, at 105 MB; and EPHEMERAL at /dev/nvme1n1p4 with xfs filesystem, encryption disabled, at 2.0 TB shown as the dominant blue bar in the visualization. A color-coded legend identifies each partition type. The Sidero Omni management plane confirms the replacement NVMe is detected and unallocated on Talos Linux node op1, ready for Ceph to claim as a new OSD. The existing 2 TB boot drive shows Talos's characteristic minimal partition scheme with tiny EFI, META, and STATE partitions alongside the large ephemeral partition, reflecting Talos's immutable OS design where the system runs from a read-only image and only ephemeral workload data persists.
A photograph under magenta LED lighting shows the rear view of multiple Dell small form factor workstations stacked vertically, their back panels dense with cables connected to a white network switch with SFP+ ports visible across the top. Ethernet cables in red, white, and gray run between the nodes and switch, while blue and green LED indicators glow through the tangle of power and data cables. Each chassis shows ventilation grilles, USB ports, HDMI outputs, and DisplayPort connections. A sticker and other desk items are partially visible on the shelf above. The three Ceph storage nodes with freshly installed WD_BLACK NVMe drives are wired back into the homelab cluster and powering on, connected via the SFP+ capable switch for the high-bandwidth interconnect that distributed storage demands. The beautiful chaos of cables reflects the reality of homelab infrastructure where enterprise-grade Kubernetes and Ceph deployments live alongside desk clutter, with fingers crossed that the OSDs rebuild cleanly and placement groups return to active plus clean status.
A photograph bathed in magenta lighting shows a hand inserting a WD_BLACK NVMe SSD into an open Dell small form factor workstation. The compact chassis reveals densely packed internals including a second NVMe already seated in an M.2 slot, SATA cables, an SK hynix branded component, and a traditional spinning hard drive visible in the drive bay. The Dell branding is partially visible on the chassis edge. This captures the hands-on hardware work of installing replacement NVMe drives into one of three Ceph storage cluster nodes after a drive failure took out the distributed storage system. The small form factor Dell machines serve as compact Kubernetes nodes running Talos Linux, packing enterprise storage capability into desktop-class hardware for a homelab Ceph deployment providing CephFS, RBD block, and S3 object storage.
Three WD_BLACK SN850X NVMe SSDs fanned out on a grid cutting mat under vivid magenta LED lighting. Two drives show the black heatsink label with white branding, the third is flipped revealing its green PCB and gold M.2 connector pins. A white mechanical keyboard with green backlighting sits in the upper left, a grapefruit tequila seltzer can in the upper right, and a white desk lamp base along the right edge. Replacement drives staged for installation into three Ceph storage cluster nodes after a failed NVMe cascaded into a cluster outage. The SN850X's PCIe Gen4 speeds suit the distributed storage workload across Talos Linux and Kubernetes. The tequila seltzer and aesthetic purple lighting capture the vibe of late-night infrastructure recovery, hardware prep before seating drives, rebuilding OSDs, and waiting for Ceph to rebalance.
https://www.anthropic.com/research/claude-code-expertise
GitHub Gist by usrbinkat containing README.md for "Public IP Address Mapping in Multi-Rack Kubernetes Clusters," showing the 18-section table of contents covering five implementation tiers ranked by Kubernetes/Cilium BGP idiomaticity for resolving Public IP → Instance → Node → ToR Switch mappings on bare-metal clusters. https://gist.github.com/usrbinkat/6bb637b08212b4d9090cea011e276d3c#decision-matrix
A screenshot shows terminal output from the Open Sesame CLI demonstrating batch repository cloning functionality. The dark-themed terminal displays tabs for bash and ssh with username usrbinkat-optiplexprime and timestamps showing 10:51. The command sesame clone --project https://github.com/ScopeCreep-zip executes from the open-sesame main branch on host mithril. The output reports Found 11 repositories in github.com/ScopeCreep-zip followed by a vertical list of repository names with existence status. Each repository shows in muted gray text with "(exists)" annotation: website, Rack, kalilix, ossiodata, dollhaus.zip, open-sesame, Symbiont, SpiritStream, ReKindle, and Springtale. The summary line reports 0 cloned, 10 skipped (exist), 0 failed, indicating all discovered repositories were already present locally and no new clones were necessary. The prompt returns to usrbinkat@mithril open-sesame main ready for the next command. This demonstrates Open Sesame's project-level batch cloning capability that discovers all repositories under a GitHub organization or user URL and clones them in a single operation. The exists detection prevents duplicate cloning by checking local workspace before initiating downloads, enabling efficient workspace synchronization across machines or after system reinstalls. This workflow automation eliminates manual navigation to individual repository pages and repetitive git clone commands, replacing dozens of manual operations with a single command that handles organization-wide repository management intelligently.
A screenshot shows terminal output from the Open Sesame CLI running workspace management commands in a dark-themed terminal. The top bar displays bash and ssh tabs with username usrbinkat-optiplexprime and timestamp 08:31. The command sesame workspace list followed by sesame workspace status show their outputs in monospace font with syntax highlighting. The workspace list displays a hierarchical tree of 20 git repositories organized under 12 organizations across 3 servers. Each server section begins with a URL, followed by indented organizations and repositories. Repositories show name, branch, abbreviated commit hash, and status in green for clean or yellow for dirty. Workspaces are tagged with workspace annotations. Projects include infrastructure and k9 under git.braincraft.io/braincraft, gitoxide under GitoxideLabs, nix and nixpkgs under NixOS, comfy-table under Nukesor, and Springtale, open-sesame marked dirty, rekindle, and spiritstream under github.com/ScopeCreep-zip. Additional repositories span catppuccin, nix-community, siderolabs, wiedymi, qemu-project, and virtio-fs. Summary reads 3 servers, 12 orgs, 20 repos. The workspace status shows open-sesame repository details: path /workspace/usrbinkat/github.com/ScopeCreep-zip/open-sesame, remote URL, branch main, commit 5662337 with message chore release 1.6.19 skip ci, tracking status origin/main up to date, dirty working tree, no profile, and namespace ScopeCreep-zip with workspace.git indicator. This demonstrates Open Sesame's workspace management functionality beyond window switching: comprehensive git repository oversight across multiple hosting platforms, providing at-a-glance status of entire project ecosystems with branch tracking and uncommitted change detection for developers managing dozens of repositories simultaneously.
selfie of usrbinkat touching grass (aka outside by the river)
DHS AND ICE ARE FEDERAL TERRORIST ORGANIZATIONS
Screenshot of a music streaming app. Text reads “Claude’s Girl” — Marika Hackman, with a small green checkmark icon. The cover art is a grainy, blue-toned photo of a sparse room: a dark, striped mattress sits against a blue wall; a person in a yellow top and gray patterned skirt lies sprawled across it with their face hidden, one bare leg dangling over the edge. A small black bedside table sits to the left. https://open.spotify.com/track/2RaUxxkie3gGE385mTEODv
A screenshot shows a GitHub repository page for folke/snacks.nvim displaying the file snacks.lua in the plugin directory. The dark-themed interface shows the repository navigation with Code tab active, Issues showing 44, and Pull requests showing 37. The breadcrumb navigation displays snacks.nvim / plugin / snacks.lua with the branch selector showing main. A commit by folke from 2 years ago appears below the file path. The file statistics show 1 lines of code, 1 loc, and 18 Bytes. The code viewer displays tabs for Code and Blame, with the Code tab active. Line 1 contains a single statement: require("snacks") written in Lua syntax with the function name in blue and the string argument in orange quotes. The minimalist file demonstrates extreme brevity, consisting solely of loading the snacks module without any additional configuration, documentation, or implementation code. This repository represents snacks.nvim, a Neovim plugin by folke, a prominent figure in the Neovim ecosystem. The humorous minimalism of the plugin file reducing the entire entry point to a single require statement plays on programming culture's appreciation for conciseness while punning on the dual meaning of require as both a Lua module loading function and a declaration of necessity.