MinIO delete-marker convergence: how the Silo fork fixed the 405 replication loop
The 405 storm had three root defects in the delete-marker path. PR #162 fixed the first attempt; PR #184 fixes retries, purge status and MRF healing in the Silo fork.
The 405 storm had three root defects in the delete-marker path. PR #162 fixed the first attempt; PR #184 fixes retries, purge status and MRF healing in the Silo fork.
MinIO Community Edition has had no release since October 2025 and the repository is archived. Three CVEs will never be patched there. Here is how to assess what your cluster is actually exposed to, and where the maintained community fork fits in.
Two movements meet: AI is getting production responsibilities, and production data is flowing through AI. Anthropic’s September report, malicious routers, training data provenance and model drift show what to verify before an AI reaches production.
HEAD probes on delete markers answer 405 and replication counts that as success. 20M stuck versions turned the loop into 684M useless requests in 48 h.
A preregistered ETH Zurich study found that computer science knowledge, not writing skill, is the strongest predictor of vibe coding success. Here is what that does and does not say about DHH’s claim that programmers can be worse at AI coding than people who cannot code.
GPT-6-Astra XHIGH sits alone on the DeepSWE cost-performance frontier, and an older model beats a flagship tier. Follow-up: DeepSeek V4.1 Flash hits 98% of Astra’s score at 1.4% of the cost.
During a MinIO replication incident three of the signals you reach for mislead you. The 405 on HeadObject is the success path, a retried 429 is invisible, and worker queue overflow logs nothing under the default priority. The counter that tracks dropped objects is incremented in four places and read in none.
AI products moved from the answer arrives to the answer flows. What streaming inference changes in capacity planning: TTFT, token throughput, concurrent streams.
DHH told Lex Fridman that for some problems, programmers do worse with AI coding tools than people who cannot code. What the claim actually covers, and what it means for teams adopting AI agents.
Part 1 of my benchmark best practices series: how I turn a scaling claim into an implied serial fraction with Amdahl’s law, and what the Universal Scalability Law adds. Includes a measured pipeline where assigning more vCPUs per image cut global throughput.