{"id":59758,"date":"2026-11-12T13:01:29","date_gmt":"2026-11-12T03:01:29","guid":{"rendered":"https:\/\/www.cloudproinc.com.au\/index.php\/2026\/11\/12\/how-dacs-reduces-ai-startup-delays-and-gpu-waste-on-azure-aks\/"},"modified":"2026-11-12T13:02:58","modified_gmt":"2026-11-12T03:02:58","slug":"how-dacs-reduces-ai-startup-delays-and-gpu-waste-on-azure-aks","status":"publish","type":"post","link":"https:\/\/www.cloudproinc.com.au\/index.php\/2026\/11\/12\/how-dacs-reduces-ai-startup-delays-and-gpu-waste-on-azure-aks\/","title":{"rendered":"How DACS Reduces AI Startup Delays and GPU Waste on Azure AKS"},"content":{"rendered":"\n<p class=\"wp-block-paragraph\">In this blog post How DACS Reduces AI Startup Delays and GPU Waste on Azure AKS we will explain how Microsoft\u2019s new storage accelerator works, why platform teams should care, and what to assess before starting a pilot.<\/p>\n\n\n\n<!--more-->\n\n\n\n<p class=\"wp-block-paragraph\">The problem is easy to miss. Your organisation invests in expensive graphics processing units, or GPUs, to run AI models, but those GPUs can spend valuable time waiting for model files and datasets to arrive from cloud storage.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">This delay becomes more noticeable when an AI service scales out, restarts after a failure or deploys a new model version. Each new workload may download the same large files again, increasing startup time, network traffic and storage requests while paid compute sits idle.<\/p>\n\n\n\n<h2 class=\"wp-block-heading\">What is the Distributed Accelerator for Cloud Storage?<\/h2>\n\n\n\n<p class=\"wp-block-paragraph\">The Distributed Accelerator for Cloud Storage, known as DACS, is a distributed read-through cache for AI workloads running on Azure Kubernetes Service. AKS is Microsoft\u2019s managed platform for running and scaling container-based applications.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">In plain English, DACS places a fast temporary storage layer between an AI application and Azure Blob Storage. Blob Storage remains the permanent, authoritative home for the model, while frequently requested data is copied onto fast local NVMe disks inside the AKS cluster.<\/p>\n\n\n\n<pre class=\"wp-block-code\"><code>AI workload requests model data\n \u2193\nIs the data already in the DACS cache?\n \u2193 \u2193\n Yes No\n \u2193 \u2193\nRead from local NVMe Read from Azure Blob Storage\n \u2193\n Add a copy to the cache\n \u2193\n Reuse it on later requests<\/code><\/pre>\n\n\n\n<p class=\"wp-block-paragraph\">This is called a read-through cache because missing data is retrieved from Blob Storage and then made available for later reads. The application can continue using Blob Storage when the requested data is not cached or the cache is unavailable.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">DACS is not a replacement for Blob Storage, a backup system or permanent storage. Its job is to make repeated reads faster and reduce how often the same information must travel across the Azure network.<\/p>\n\n\n\n<h2 class=\"wp-block-heading\">Why platform teams should pay attention<\/h2>\n\n\n\n<h3 class=\"wp-block-heading\">1. It can reduce the time GPUs spend waiting<\/h3>\n\n\n\n<p class=\"wp-block-paragraph\">Large AI models may contain tens or hundreds of gigabytes of model weights\u2014the numerical information the model must load before it can answer a request. If every new AI pod, meaning a running application instance, downloads those files from remote storage, startup can be slow.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">DACS allows multiple workloads to reuse model data already brought into the cluster. The potential business outcome is faster service recovery, more responsive scaling and better use of costly GPU capacity.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">This does not mean every workload will become faster. The first read still comes from Blob Storage, and the result depends on how often data is reused, the size of the active models and whether the cache is already warm.<\/p>\n\n\n\n<h3 class=\"wp-block-heading\">2. Scaling becomes less dependent on repeated downloads<\/h3>\n\n\n\n<p class=\"wp-block-paragraph\">AI demand is rarely constant. An internal assistant might be quiet overnight and experience a surge when employees begin work, while a customer-facing service can change rapidly after a product launch.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">Without caching, adding GPU nodes can trigger several copies of the same model to be downloaded at once. DACS gives replacement and scale-out workloads an opportunity to reuse data already cached across the cluster, reducing pressure on Blob Storage and the network.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">This can also help during rolling upgrades, where application instances are gradually replaced. It does not eliminate deployment planning, but it may remove one common source of delay.<\/p>\n\n\n\n<h3 class=\"wp-block-heading\">3. Storage architecture becomes a cost decision<\/h3>\n\n\n\n<p class=\"wp-block-paragraph\">DACS uses local NVMe storage, which is very fast storage attached to Azure virtual machines. Platform teams therefore need to compare the cost of dedicated cache nodes with the value of shorter startup times, fewer repeated storage reads and less wasted GPU time.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">The cache should be sized for the active working set\u2014the models and dataset portions being reused now\u2014not every AI file the organisation owns. Additional room may also be needed when old and new model versions overlap during a release.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">The right question is not simply, \u201cIs the cache fast?\u201d It is, \u201cDoes the time and compute capacity it saves justify the extra infrastructure and operational work?\u201d<\/p>\n\n\n\n<h2 class=\"wp-block-heading\">What changes for the platform team?<\/h2>\n\n\n\n<p class=\"wp-block-paragraph\">DACS is managed through Azure Container Storage and currently requires a suitable Linux node pool with local NVMe disks. Workloads then access the cache through supported options such as the native Python client, BlobFuse2 or Run:ai Model Streamer.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">BlobFuse2 is useful when an application expects cloud files to look like a normal Linux file system. The native client provides more direct application integration, while Model Streamer is aimed at loading model data efficiently into GPU memory.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">That means DACS is not a switch that can be enabled without planning. Teams need to consider:<\/p>\n\n\n\n<ul class=\"wp-block-list\">\n<li><strong>Workload compatibility:<\/strong> How the AI application currently reads models and datasets.<\/li>\n<li><strong>Capacity:<\/strong> How much model data is actively reused and how much local disk is available.<\/li>\n<li><strong>Cold starts:<\/strong> How the service behaves before the cache has been populated.<\/li>\n<li><strong>Prefetching:<\/strong> Whether a new model should be loaded into the cache before production traffic arrives.<\/li>\n<li><strong>Failure testing:<\/strong> What happens when cache nodes are replaced or temporarily unavailable.<\/li>\n<li><strong>Monitoring:<\/strong> How cache hits, latency, capacity, network pressure and Blob Storage fallback will be measured.<\/li>\n<\/ul>\n\n\n\n<p class=\"wp-block-paragraph\">DACS exposes Prometheus-format operational metrics, but the preview does not automatically connect them to Azure Monitor managed Prometheus. Platform teams must configure collection and alerting rather than assuming monitoring is included.<\/p>\n\n\n\n<h2 class=\"wp-block-heading\">A practical scenario<\/h2>\n\n\n\n<p class=\"wp-block-paragraph\">Consider a 200-person software company operating an AI support assistant on AKS. Its inference service runs six GPU-backed instances, each of which downloads the same model files during deployment.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">A routine update replaces all six instances. The model is downloaded repeatedly, deployment takes longer than expected and GPUs remain allocated while the service prepares to accept requests.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">With DACS, the first workload can retrieve the required data from Blob Storage and populate the shared cache. Later workloads can reuse the cached chunks rather than independently retrieving every file again.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">The company would measure the result using four business-focused indicators: model startup time, GPU idle minutes, Blob Storage read volume and recovery time after a node failure. If those figures do not improve enough to justify the cache nodes and management effort, DACS may not be the right investment.<\/p>\n\n\n\n<h2 class=\"wp-block-heading\">The security limitation leaders should not overlook<\/h2>\n\n\n\n<p class=\"wp-block-paragraph\">DACS currently treats the whole Kubernetes cluster as one trusted environment. Any workload that can reach a DACS cache endpoint is considered trusted, and the preview does not provide isolation between separate tenants or untrusted workloads.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">This makes cluster design important. Sensitive models should not be placed in a shared cluster where unrelated or less trusted applications can access the cache network.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">Identity controls are still required for Blob Storage, along with network restrictions, tightly controlled Kubernetes permissions and clear ownership of cache administration. DACS improves data delivery; it does not replace the security controls discussed in our guide to production AI agent security on AKS.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">It should also be considered alongside the access controls used when connecting AI agents to Azure Blob Storage. Faster access to data must never mean broader or poorly governed access.<\/p>\n\n\n\n<h2 class=\"wp-block-heading\">When DACS is not the right answer<\/h2>\n\n\n\n<p class=\"wp-block-paragraph\">DACS is currently a public preview without a service-level agreement, so it should be evaluated carefully rather than treated as a default production component. Microsoft does not currently recommend preview features for critical production workloads.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">It is also unlikely to help when data is read only once, models are small, workloads rarely restart or the cache costs more than the delays it removes.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">The preview is not designed for permanent training outputs, checkpoints, general shared-file storage, container image caching, large language model conversation memory or storage platforms other than Azure Blob Storage.<\/p>\n\n\n\n<h2 class=\"wp-block-heading\">How to approach a DACS pilot<\/h2>\n\n\n\n<ol class=\"wp-block-list\">\n<li>Choose one non-critical AI workload with large, repeatedly read model files.<\/li>\n<li>Record current startup time, GPU idle time, storage reads and recovery performance.<\/li>\n<li>Deploy a small, isolated DACS cache using representative data.<\/li>\n<li>Test cold-cache, warm-cache, scale-out, upgrade and node-failure scenarios.<\/li>\n<li>Review security boundaries and confirm that every workload in the cluster is trusted.<\/li>\n<li>Compare measurable savings with cache-node and support costs before expanding.<\/li>\n<\/ol>\n\n\n\n<p class=\"wp-block-paragraph\">DACS addresses a genuine problem: expensive AI compute often waits for data. For the right AKS workload, keeping frequently used model data close to GPUs could shorten startup times, reduce repeated storage traffic and help platform teams get more value from their infrastructure.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">The preview limitations mean it is not yet a universal production answer. A controlled pilot, backed by clear performance and cost measurements, is the sensible next step.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">CloudPro Inc combines Microsoft Partner and Wiz Security Integrator experience with more than 20 years of enterprise IT delivery. If you are unsure whether storage delays are wasting GPU capacity in your Azure AI environment, our Melbourne-based team can help assess the architecture and pilot options\u2014no strings attached.<\/p>\n\n\n","protected":false},"excerpt":{"rendered":"<p>DACS keeps frequently used AI model data close to Azure GPUs. Learn where it could reduce startup delays, storage traffic and wasted compute\u2014and where its preview limitations matter.<\/p>\n","protected":false},"author":1,"featured_media":59760,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"_et_pb_use_builder":"","_et_pb_old_content":"","_et_gb_content_width":"","_jetpack_newsletter_access":"","_jetpack_dont_email_post_to_subs":true,"_jetpack_newsletter_tier_id":0,"_jetpack_memberships_contains_paywalled_content":false,"_jetpack_hide_featured_image":false,"_jetpack_featured_image_duplicate_dismissed":0,"_jetpack_feature_clip_id":0,"_jetpack_memberships_contains_paid_content":false,"footnotes":"","jetpack_post_was_ever_published":false},"categories":[24,16,139,13],"tags":[],"class_list":["post-59758","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-ai","category-microsoft-azure","category-azure-kubernetes-service","category-blog"],"jetpack-related-posts":[{"id":59774,"url":"https:\/\/www.cloudproinc.com.au\/index.php\/2026\/11\/13\/running-ai-workloads-on-aks-with-distributed-storage-caching\/","url_meta":{"origin":59758,"position":0},"title":"Running AI Workloads on AKS with Distributed Storage Caching","author":"CPI Staff","date":"November 13, 2026","format":false,"excerpt":"Distributed storage caching can reduce AI startup delays and wasted GPU time on AKS. Here is how it works, where it helps and what leaders should assess before investing.","rel":"","context":"In &quot;AI&quot;","block_context":{"text":"AI","link":"https:\/\/www.cloudproinc.com.au\/index.php\/category\/ai\/"},"img":{"alt_text":"","src":"\/wp-content\/uploads\/2026\/11\/running-ai-workloads-on-aks-with-distributed-storage-caching.png","width":350,"height":200,"srcset":"\/wp-content\/uploads\/2026\/11\/running-ai-workloads-on-aks-with-distributed-storage-caching.png 1x, \/wp-content\/uploads\/2026\/11\/running-ai-workloads-on-aks-with-distributed-storage-caching.png 1.5x, \/wp-content\/uploads\/2026\/11\/running-ai-workloads-on-aks-with-distributed-storage-caching.png 2x, \/wp-content\/uploads\/2026\/11\/running-ai-workloads-on-aks-with-distributed-storage-caching.png 3x, \/wp-content\/uploads\/2026\/11\/running-ai-workloads-on-aks-with-distributed-storage-caching.png 4x"},"classes":[]},{"id":59520,"url":"https:\/\/www.cloudproinc.com.au\/index.php\/2026\/11\/03\/how-aks-flex-nodes-extend-azure-clusters-to-your-own-compute\/","url_meta":{"origin":59758,"position":1},"title":"How AKS Flex Nodes Extend Azure Clusters to Your Own Compute","author":"CPI Staff","date":"November 3, 2026","format":false,"excerpt":"AKS Flex Nodes connect your own servers and virtual machines to an Azure Kubernetes cluster. Learn where the preview could reduce costs and where it adds risk.","rel":"","context":"In &quot;Azure&quot;","block_context":{"text":"Azure","link":"https:\/\/www.cloudproinc.com.au\/index.php\/category\/microsoft-azure\/"},"img":{"alt_text":"","src":"\/wp-content\/uploads\/2026\/11\/how-aks-flex-nodes-extend-azure-clusters-to-your-own-compute.png","width":350,"height":200,"srcset":"\/wp-content\/uploads\/2026\/11\/how-aks-flex-nodes-extend-azure-clusters-to-your-own-compute.png 1x, \/wp-content\/uploads\/2026\/11\/how-aks-flex-nodes-extend-azure-clusters-to-your-own-compute.png 1.5x, \/wp-content\/uploads\/2026\/11\/how-aks-flex-nodes-extend-azure-clusters-to-your-own-compute.png 2x, \/wp-content\/uploads\/2026\/11\/how-aks-flex-nodes-extend-azure-clusters-to-your-own-compute.png 3x, \/wp-content\/uploads\/2026\/11\/how-aks-flex-nodes-extend-azure-clusters-to-your-own-compute.png 4x"},"classes":[]},{"id":58761,"url":"https:\/\/www.cloudproinc.com.au\/index.php\/2026\/10\/12\/connecting-openai-agents-to-azure-blob-storage-and-cloud-data\/","url_meta":{"origin":59758,"position":2},"title":"Connecting OpenAI Agents to Azure Blob Storage and Cloud Data","author":"CPI Staff","date":"October 12, 2026","format":false,"excerpt":"Learn how to give OpenAI agents controlled access to Azure Blob Storage and cloud data without exposing sensitive files, creating security gaps, or building an expensive data platform.","rel":"","context":"In &quot;AI Agents&quot;","block_context":{"text":"AI Agents","link":"https:\/\/www.cloudproinc.com.au\/index.php\/category\/ai-agents\/"},"img":{"alt_text":"","src":"\/wp-content\/uploads\/2026\/09\/connecting-openai-agents-to-azure-blob-storage-and-cloud-data.png","width":350,"height":200,"srcset":"\/wp-content\/uploads\/2026\/09\/connecting-openai-agents-to-azure-blob-storage-and-cloud-data.png 1x, \/wp-content\/uploads\/2026\/09\/connecting-openai-agents-to-azure-blob-storage-and-cloud-data.png 1.5x, \/wp-content\/uploads\/2026\/09\/connecting-openai-agents-to-azure-blob-storage-and-cloud-data.png 2x, \/wp-content\/uploads\/2026\/09\/connecting-openai-agents-to-azure-blob-storage-and-cloud-data.png 3x, \/wp-content\/uploads\/2026\/09\/connecting-openai-agents-to-azure-blob-storage-and-cloud-data.png 4x"},"classes":[]},{"id":59598,"url":"https:\/\/www.cloudproinc.com.au\/index.php\/2026\/11\/06\/how-to-secure-mcp-servers-on-aks-with-a-governed-ai-agent-gateway\/","url_meta":{"origin":59758,"position":3},"title":"How to Secure MCP Servers on AKS with a Governed AI Agent Gateway","author":"CPI Staff","date":"November 6, 2026","format":false,"excerpt":"Learn how Application Gateway for Containers can give AKS-hosted MCP servers one governed entry point, reducing exposure and improving control over production AI agents.","rel":"","context":"In &quot;Azure Kubernetes Service&quot;","block_context":{"text":"Azure Kubernetes Service","link":"https:\/\/www.cloudproinc.com.au\/index.php\/category\/azure-kubernetes-service\/"},"img":{"alt_text":"","src":"\/wp-content\/uploads\/2026\/11\/how-to-secure-mcp-servers-on-aks-with-a-governed-ai-agent-gateway.png","width":350,"height":200,"srcset":"\/wp-content\/uploads\/2026\/11\/how-to-secure-mcp-servers-on-aks-with-a-governed-ai-agent-gateway.png 1x, \/wp-content\/uploads\/2026\/11\/how-to-secure-mcp-servers-on-aks-with-a-governed-ai-agent-gateway.png 1.5x, \/wp-content\/uploads\/2026\/11\/how-to-secure-mcp-servers-on-aks-with-a-governed-ai-agent-gateway.png 2x, \/wp-content\/uploads\/2026\/11\/how-to-secure-mcp-servers-on-aks-with-a-governed-ai-agent-gateway.png 3x, \/wp-content\/uploads\/2026\/11\/how-to-secure-mcp-servers-on-aks-with-a-governed-ai-agent-gateway.png 4x"},"classes":[]},{"id":59316,"url":"https:\/\/www.cloudproinc.com.au\/index.php\/2026\/10\/22\/how-azure-monitor-advanced-metrics-reduce-risk-and-cloud-costs\/","url_meta":{"origin":59758,"position":4},"title":"How Azure Monitor Advanced Metrics Reduce Risk and Cloud Costs","author":"CPI Staff","date":"October 22, 2026","format":false,"excerpt":"Advanced platform metrics reveal where Azure growth is occurring, helping IT leaders investigate unusual activity sooner, assign ownership and control costs before the monthly invoice arrives.","rel":"","context":"In &quot;Azure Monitor&quot;","block_context":{"text":"Azure Monitor","link":"https:\/\/www.cloudproinc.com.au\/index.php\/category\/azure-monitor\/"},"img":{"alt_text":"","src":"\/wp-content\/uploads\/2026\/10\/how-azure-monitor-advanced-metrics-reduce-risk-and-cloud-costs.png","width":350,"height":200,"srcset":"\/wp-content\/uploads\/2026\/10\/how-azure-monitor-advanced-metrics-reduce-risk-and-cloud-costs.png 1x, \/wp-content\/uploads\/2026\/10\/how-azure-monitor-advanced-metrics-reduce-risk-and-cloud-costs.png 1.5x, \/wp-content\/uploads\/2026\/10\/how-azure-monitor-advanced-metrics-reduce-risk-and-cloud-costs.png 2x, \/wp-content\/uploads\/2026\/10\/how-azure-monitor-advanced-metrics-reduce-risk-and-cloud-costs.png 3x, \/wp-content\/uploads\/2026\/10\/how-azure-monitor-advanced-metrics-reduce-risk-and-cloud-costs.png 4x"},"classes":[]},{"id":435,"url":"https:\/\/www.cloudproinc.com.au\/index.php\/2026\/10\/12\/how-to-deploy-a-wiz-outpost-to-microsoft-azure\/","url_meta":{"origin":59758,"position":5},"title":"How to Deploy a Wiz Outpost to Microsoft Azure","author":"CPI Staff","date":"October 12, 2026","format":false,"excerpt":"This Wiz Security article will explain how to deploy a Wiz outpost to Microsoft Azure cloud infrastructure. Listen to this article Before we discuss how to deploy an outpost, it is important to understand why many companies do so. A Wiz outpost is the deployment of the Wiz security scanner\u2026","rel":"","context":"In &quot;Azure&quot;","block_context":{"text":"Azure","link":"https:\/\/www.cloudproinc.com.au\/index.php\/category\/microsoft-azure\/"},"img":{"alt_text":"","src":"\/wp-content\/uploads\/2024\/07\/How-to-Deploy-a-Wiz-Outpost-to-Microsoft-Azure.webp","width":350,"height":200,"srcset":"\/wp-content\/uploads\/2024\/07\/How-to-Deploy-a-Wiz-Outpost-to-Microsoft-Azure.webp 1x, \/wp-content\/uploads\/2024\/07\/How-to-Deploy-a-Wiz-Outpost-to-Microsoft-Azure.webp 1.5x, \/wp-content\/uploads\/2024\/07\/How-to-Deploy-a-Wiz-Outpost-to-Microsoft-Azure.webp 2x, \/wp-content\/uploads\/2024\/07\/How-to-Deploy-a-Wiz-Outpost-to-Microsoft-Azure.webp 3x, \/wp-content\/uploads\/2024\/07\/How-to-Deploy-a-Wiz-Outpost-to-Microsoft-Azure.webp 4x"},"classes":[]}],"jetpack_sharing_enabled":true,"jetpack_featured_media_url":"\/wp-content\/uploads\/2026\/11\/how-dacs-reduces-ai-startup-delays-and-gpu-waste-on-azure-aks.png","_links":{"self":[{"href":"https:\/\/www.cloudproinc.com.au\/index.php\/wp-json\/wp\/v2\/posts\/59758","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/www.cloudproinc.com.au\/index.php\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/www.cloudproinc.com.au\/index.php\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/www.cloudproinc.com.au\/index.php\/wp-json\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"https:\/\/www.cloudproinc.com.au\/index.php\/wp-json\/wp\/v2\/comments?post=59758"}],"version-history":[{"count":2,"href":"https:\/\/www.cloudproinc.com.au\/index.php\/wp-json\/wp\/v2\/posts\/59758\/revisions"}],"predecessor-version":[{"id":59761,"href":"https:\/\/www.cloudproinc.com.au\/index.php\/wp-json\/wp\/v2\/posts\/59758\/revisions\/59761"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/www.cloudproinc.com.au\/index.php\/wp-json\/wp\/v2\/media\/59760"}],"wp:attachment":[{"href":"https:\/\/www.cloudproinc.com.au\/index.php\/wp-json\/wp\/v2\/media?parent=59758"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/www.cloudproinc.com.au\/index.php\/wp-json\/wp\/v2\/categories?post=59758"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/www.cloudproinc.com.au\/index.php\/wp-json\/wp\/v2\/tags?post=59758"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}