OpenLink Software’s Post

🚀 Strengthening API Security in ORDS with OAuth 2.0 🔐 Just published my latest blog on securing APIs in ORDS using OAuth! In this article, I’ve explained how we can move from traditional Basic Authentication to a more secure and scalable OAuth 2.0 approach in ORDS. I’ve covered everything from client creation, role and privilege management, token generation, and how ORDS validates access behind the scenes. If you're working with ORDS APIs or exploring API security in Oracle environments, this will give you a clear and practical understanding of how OAuth actually works in real scenarios. 📖 Read here: https://lnkd.in/giPJenwn Would love to hear your thoughts and feedback! #Oracle #ORDS #OAuth #APISecurity #Database #OracleDatabase #BackendDevelopment #RESTAPI #Tech #Developers #CloudSecurity #Learning #DBMS

𝐒𝐲𝐛𝐚𝐬𝐞 𝐀𝐒𝐄 𝐢𝐬 𝐬𝐭𝐢𝐥𝐥 𝐫𝐮𝐧𝐧𝐢𝐧𝐠 𝐜𝐨𝐫𝐞 𝐞𝐧𝐭𝐞𝐫𝐩𝐫𝐢𝐬𝐞 𝐰𝐨𝐫𝐤𝐥𝐨𝐚𝐝𝐬. But stability and performance don’t happen by default.   𝐈𝐭 𝐫𝐞𝐪𝐮𝐢𝐫𝐞𝐬 𝐭𝐡𝐞 𝐫𝐢𝐠𝐡𝐭 𝐞𝐧𝐠𝐢𝐧𝐞𝐞𝐫𝐢𝐧𝐠 𝐚𝐩𝐩𝐫𝐨𝐚𝐜𝐡. At OptiSol, we bring AI-driven insights into Sybase environments—ensuring every layer is tuned for reliability, performance, and continuity. 𝐄𝐱𝐩𝐥𝐨𝐫𝐞 𝐦𝐨𝐫𝐞 :https://lnkd.in/gaSuPgqr 👉 𝐄𝐱𝐩𝐥𝐨𝐫𝐞 𝐨𝐮𝐫 𝐚𝐩𝐩𝐫𝐨𝐚𝐜𝐡 🔹 𝐒𝐜𝐡𝐞𝐦𝐚 & 𝐂𝐨𝐝𝐞 Structured, ordered releases with compatibility for existing applications 🔹 𝐏𝐥𝐚𝐧𝐬 & 𝐂𝐨𝐧𝐜𝐮𝐫𝐫𝐞𝐧𝐜𝐲 Indexes, locking, and isolation tuned to real workloads—not one-size defaults 🔹 𝐄𝐧𝐠𝐢𝐧𝐞 𝐎𝐩𝐭𝐢𝐦𝐢𝐳𝐚𝐭𝐢𝐨𝐧 Cache, parallelism, and tempdb sized for peak usage 🔹 𝐂𝐥𝐢𝐞𝐧𝐭𝐬 & 𝐃𝐫𝐢𝐯𝐞𝐫𝐬 JDBC, ODBC, PowerBuilder validated under real load 🔹 𝐑𝐞𝐩𝐥𝐢𝐜𝐚𝐭𝐢𝐨𝐧 & 𝐃𝐑 Replication and standby paths always in sync with releases 🔹 𝐏𝐫𝐨𝐝𝐮𝐜𝐭𝐢𝐨𝐧 𝐎𝐩𝐞𝐫𝐚𝐭𝐢𝐨𝐧𝐬 Monitoring, alerts, and runbooks that clearly define system health   𝑾𝒉𝒚 𝒊𝒕 𝒎𝒂𝒕𝒕𝒆𝒓: ✔ Reduced performance bottlenecks ✔ Stable, compliant operations ✔ Predictable, disruption-free releases Because in enterprise systems, stability isn’t accidental—it’s engineered. 𝑹𝒆𝒂𝒄𝒉 𝒐𝒖𝒕 𝐨𝐩𝐭𝐢𝐬𝐨𝐥 - 📧 solutions@optisolbusiness.com 📞 +1 412 406 9010   #Sybase #SybaseASE #ASE #DBA #DatabasePerformance #Replication #Sybase #AdaptiveServerEnterprise #SybaseSupport #DatabaseSupport #DatabaseAdministration #LegacyDatabase #DatabaseMigration #SybaseMigration #PostgreSQLMigration #LegacyModernization #DataEngineering #ReplicationServer #ETL #EnterpriseIT #ITOperations #ITInfrastructure #DigitalTransformation #ITStrategy #DatabaseManagement #DatabaseExperts #GDPRCertified #SOCCertified #EnterpriseIT #DatabaseEngineering #DataPlatforms

Encryption for SQL Server is no longer a best practice; it’s a baseline requirement. Let’s walk through what’s changing, why it matters, and how you should adjust your environments. #SQLServer #AzureSQL

There's a category of IT problem I'd describe as "someone else's strategic decision becoming your operational emergency." Oracle's decision not to port Rdb to OpenVMS x86_64 is one of those. The organizations affected didn't choose this situation. They chose to run stable, long-lived systems that worked. Now they're being handed a platform constraint with no obvious exit. We've been thinking about this for years. Happy to talk to anyone navigating it. #OpenVMS #OracleRdb #EnterpriseIT #LegacyModernization #MimerSQL #DatabaseMigration

A bug in SSMS v22.5.0? So, heres the thing. I have a small test network at home and access a cluster remotely from an admin server running Windows Server 2025 that has my software on it. Part of this software is SSMS 22.5.0, freshly installed. It connects to a cluster with Windows Server 2025 and SQL Server 2025. The system volume is 100GB and with around 65GB free before running the script. When I execute long running TSQL the system volume on the admin machine fills up and the query execution fails with "no space available". In the image there are 4 temporary files each around 14GB that could represent one for each core on the admin machine. My question is why I this system drive filling up with temporary files when I execute a script on a remote server? This is driving me mad! UPDATE: I can only delete these temporary files once I have closed the SSMS. That sort of points the finger at the perpetrator! UPDATE 2: DISCLAIMER: Due to an ashamedly large amount of frustration I asked Copilot what the problem may be. Here is the answer: /***********************************************/ The system drive on your admin machine is filling up because SQL Server Management Studio (SSMS) 22.5.0 caches large query results to your local TEMP directory on the C: drive before displaying them. Even though the query runs on a remote server, SSMS brings that data back over the network to the machine running the application.  Database Administrators Stack Exchange  +3 Because you are likely returning a massive dataset or the query is spooling in a way that generates enormous output, SSMS creates these large .tmp files to hold the data, resulting in the "no space available" error.  Microsoft Learn  +1 Why This Happens (The "Perpetrator") SSMS Caching Behavior: SSMS acts as a 32-bit application (even on 64-bit OSs) and cannot hold massive result sets in memory. To handle this, it writes result sets to C:\Users\<User>\AppData\Local\Temp. "No Space Available" Failure: As you noted, the files are locked until the query tab is closed or SSMS is exited, confirming they are held open by the SSMS process holding the result set. Multiple Files: The multiple files likely correspond to multiple query windows you have open, or one massive result set that was broken into chunks. /***********************************************/ Firstly: How did I not know that? Secondly: A WTF moment. The SSMS loads my large datasets into temp files on the system partition??? Thirdly: I ran exactly the same queries in SQLCMD and the "problem" seemes to be resolved. My Conclusion: The SSMS must not be installed on database servers, and most especially not production servers. Imagine this happened on a highly productive SQL Server and the system volume ran full. It also seems that the 64-bit Azure Data Studio (ADS) does not have this problem. It is just a shame that I can't do everything in ADS that I can do in SSMS. Maybe I do need to use both tools afterall.

Oracle AI World Tour NYC just gave me a lot to think about with Deep Data Security. The identity propagation story is compelling — OAuth2 tokens carrying end-user identity all the way to the database, so agents can't exceed what the actual user is authorized to see. But one question I haven't seen answered clearly: Does the agent get a new database connection per end user — or does it inject identity into a reused pooled connection? At scale, that distinction matters enormously. A new connection per user kills throughput. Session-level context switching or proxy authentication would be the practical path — but Oracle hasn't published the pooling mechanics in detail yet. If you're designing agentic systems on Oracle and this affects your architecture, I'd love to know what you've found. #Oracle #OracleAIWorldTour #AgenticAI #DatabaseSecurity #OCI

Agent is still a good place for those system engineering guys to give better software stack and architecture. We did such MCP-SD to optimize both the latency (millisecond level) and token cost (reduce by 80%) when agent interacts with MCP. Waiting for more testing!

MCP is often criticized for wasting tokens: large tool schemas and responses are loaded into the agent context, even when only a small part is needed for reasoning. Agents often reason over only a small portion of an MCP response, while many MCP responses simply pass through the agent before being sent to another MCP server. To address this, We proposed Selective Disclosure and a Direct Data Interface (DDI) between MCP servers: the agent selects and sees only the abstract fields needed for reasoning, while full body data moves through a separate data plane. We built MCP-SD and its reference implementation S2SP to explore this idea. https://mcp-sd.github.io/

🧠 How Go Runtime Handles a Network Request (Under the Hood) When we build a backend server using Go, we usually focus on handlers, routers, and business logic. But behind the scenes, something much more interesting is happening. Let’s walk through what actually happens when a client sends data to a Go server. When a client sends a request, the data first travels through the network and reaches the server’s Network Interface Card (NIC). The NIC writes this incoming data into its receive buffer and then triggers an interrupt to notify the operating system. At this point, the kernel takes control. The kernel reads the data from the NIC buffer and places it into a socket receive buffer associated with the connection. It’s important to clarify here: the kernel does not create a new socket per request—the socket already exists (created during connection setup), and incoming data is simply appended to its buffer. Now comes the interesting part. The Go server is not constantly polling for data in a blocking way. Instead, it relies on an efficient I/O mechanism like epoll (on Linux systems). The Go runtime registers file descriptors (sockets) with epoll and then waits for events. When data arrives: The kernel marks the socket as ready epoll detects this readiness epoll wakes up the waiting thread Inside the Go runtime, this is handled by a component called the network poller. This poller runs on a separate thread and listens for I/O events using epoll. When epoll signals that data is ready: The runtime identifies which goroutine was waiting on that socket That goroutine is moved from a waiting state to a runnable state Now enters the Go scheduler. The Go runtime uses its own scheduler (based on the G-M-P model): The awakened goroutine is placed into a run queue A logical processor (P) picks it up It gets executed on an OS thread (M) At this point, your handler code finally runs: It reads the request Processes it Sends a response back 🔄 What About Sending Data Back? When the server sends a response: Data is written to the socket If the socket is not ready to send (e.g., buffer full), the goroutine is parked again epoll waits for the socket to become writable The same wake-up mechanism happens again ⚙️ Key Insight You don’t manually manage: Threads Blocking I/O Event loops The Go runtime handles all of it for you. 🚀 Why This Matters This design is why Go can handle: Thousands of concurrent connections Minimal memory usage High-performance backend systems Without you writing complex async or event-driven code. 📌 Final Thought What looks like a simple HTTP handler in Go is actually powered by: OS-level networking Kernel socket buffers epoll-based event notification Go runtime scheduler 👉 All working together seamlessly.

SolarWinds DPA's built-in SQL Server Agent alert has a blind spot: it fires on every step failure — including the intentional ones you designed into your job flow control. If you've ever built a job where step 1 tries the fast path, fails by design, and step 2 picks up the alternate route… you know the noise this creates. The fix is filtering to step_id = 0 in sysjobhistory — the job-level outcome row that tells you whether the job actually failed, not just a step. I wrote up two DPA custom alert queries that do this (simple count + job names), plus a sargable CTE version for servers with large msdb databases where the row-by-row function evaluation might add up. https://lnkd.in/gQQXzQP5 #SQLServer #DBA #SolarWinds #DPA #Monitoring