These are not summaries of someone else's documentation. They are the architecture decisions we made, the failures we hit and how we fixed them, while building and operating Z-Cloud Workspace, Z-Auto and Z-EDU — the multi-tenant SaaS products (many businesses on one system, with fully isolated data) Z-SOFT builds in-house.
Adding Redis is only part of the story. Read load should be stopped at the edge, single-flight keeps the database from being buried in duplicate queries when the cache expires, and a single misplaced Cache-Control line can leak private data.
The video player sends a progress ping every 15 seconds for every learner. As the number of learners grows, these small writes can put serious pressure on the database. This article covers how we batched the writes through Redis and why the merge rule has to live in SQL, not just in application code.
Read article : Write-behind: from 20 pings down to 2 database transactionsA data leak usually does not start with a complex attack. It can come from a query missing the tenant condition, a wrongly cached route, or a permission check that lives only in the browser. This article walks through authorization, presigned URL uploads, the server clock, and the fail-open decision.
Read article : Multi-tenant SaaS security: where data actually leaksAuto scaling does not fix a suboptimal design by itself. Before adding instances, take the unnecessary work off the request path, move side work into a queue, and alert on the backlog of work instead of just watching CPU.
Read article : Load and auto scaling: what to fix before adding instancesDDoS cannot be stopped with one button. This article walks through the defense layers Z-SOFT runs: IP limits and Slowloris timeouts in OpenResty, cache at the edge, a rate limit that runs before JWT authentication, and why many counters choose fail-open.
Read article : Multi-layer DDoS defense: from infrastructure to applicationRequests are usually lost not because servers are scarce, but because a new instance takes traffic before it is ready or an old instance is stopped with work still in flight. This article goes from saturation, p99, and the readiness probe to graceful shutdown and rolling updates.
Read article : Monitoring load and scaling without downtime: what to measure, what to do?Z-SOFT designs, builds and operates high-load systems for businesses — from mapping your business processes and designing the architecture through to security review and long-term operations.
Talk to our engineers