Problemsby GlyphDatabases hit connection, locking and scaling limits
16 posts from 16 people15d activeFading, +0 this week
Posts per day
The posts behind it
POLARS_MAX_BLOCKING_THREAD_COUNT hit without being in deadlock
### Checks - [x] I have checked that this issue has not already been reported. - [x] I have confirmed this bug exists on the [latest version](https://pypi.org/project/polars/) of Polars. ### Reproducible example Occurs in part of a larger process at work - I'll see if I can get...
`to_pandas()` deadlocks with `map_elements` across threads
### Checks - [x] I have checked that this issue has not already been reported. - [x] I have confirmed this bug exists on the [latest version](https://pypi.org/project/polars/) of Polars. ### Reproducible example ### Log output ### Issue description This is a deadlock we hit when...
[Bug]: [Usage]: RTX PRO 5000 (SM120) deadlocks at 3-4W during vLLM 0.32.0 startup with 70B NVFP4 model, even with --enforce-eager
### Your current environment **Note:** The collect_env output shows `CUDA Archs: Not Set`, meaning vLLM is not picking up `FLASHINFER_CUDA_ARCH_LIST=12.0f`. Also, `TORCHINDUCTOR_COMPILE_THREADS=1` is forcing a single-threaded compile, which may be contributing to the startup...
What an e2-micro can handle: 200 concurrent AI agent sessions on 1 GB RAM
...agents: 51 s, no swap. - 200 agents: all finished, with the machine at its limit. The bottleneck wasn't CPU or the network. It was lock contention in our own save path: every save took one account-wide lock. Per-session locks, releasing the lock before waiting on the disk, and a...
[BUG]: pglite-socket: handlers that end in an error are never removed, so maxConnections fills up and every new connection is refused
### What happens In `PGLiteSocketServer.handleConnection`, a handler is removed from `this.handlers` only by its `close` listener. When a socket errors (for example `ECONNRESET`), `PGLiteSocketHandler.handleError()` calls `detach(true)`. `detach` removes the socket's `close`...
bug: 2.26.1 Rust TUI + Unified Harness: resume/continue and fresh start can deadlock after session restore (both sides idle, py-spy evidence); stale leases with dead PIDs block resume
...PID no longer exists. On this machine 73 stale leases had accumulated over two months. Workaround: delete the lock files whose `process_id` is dead from `~/.vibe/logs/session/active/`. Suggested fix: on resume, treat a lease as free when `kill(pid, 0)` fails (or the recorded...
...up flipping it: use the "FastCGI" approach as a default, and if the original PHP code doesn't work (rare), then for that app, you can choose the regular CGI approach of forking a process per request. That way you get maximum speed by default, and manually opt into safety (this...
[Bug]: Exception on a non-output TP rank is logged and the worker dequeues the next RPC, so ranks desync and collectives mispair (cross-node deadlock)
...which dequeues the next RPC. Only the output rank reports the failure to the engine. The failing rank then starts step N+1 while its peers are still in step N, and the two sides' collectives pair with the wrong ops. In our case this ended in a cross-node deadlock and, five...
Execute read-only queries using shared MySQL connection pools
Execute unqualified consumer SQL on the resolved stack database using bounded pools shared across tenants. Parent: https://github.com/grafana/grafana/issues/134293 **Implementation locations:** New proxy repository. ## Acceptance criteria - [ ] Maintain one reader connection...
persistent 5xx/504 on /auth/v1/* during business hours
Project zvusdpixljxkxlxrrauj (Pro). We're seeing persistent 5xx/504 on /auth/v1/* during business hours. Last 7 days: 1,753 5xx of which 1,289 are 504 (Auth 60s timeout). All between 06:00-15:00 UTC on weekdays; 0 on Sunday, 18 on Saturday. Worst hours: Mon 5 Oct 10:00 UTC - 181...
Document several environment variables that n8n reads but the docs do not list
### Bug Description We run n8n self-hosted in queue mode and set the variables below in production. The code reads them, but the environment-variable reference pages do not describe them. I checked against the `n8n-io/n8n` master branch and the `n8n-io/n8n-docs` repository at...
Streaming result deadlocks (all threads idle, client in Executor::WaitForTask) on a UNION ALL of a LEFT JOIN probe and an EXISTS + NOT EXISTS branch, at threads below ~40
### What happens? The query below never finishes when its result is streamed (any duckdb-python execute() followed by fetchall(), to_arrow_table() or to_arrow_reader()). The process uses no CPU. - Where it waits. The client thread waits forever in...
bug: Resuming a session whose journal ends with an unresolved tool-approval callback deadlocks the TUI
...(~15 minutes in my case): typed input appears accepted but nothing happens, and there is no way to resolve or cancel the pending approval from the TUI. The process also ignores `SIGTERM` in this state; `SIGINT` (in-terminal `Ctrl+C` or `kill -INT `) exits cleanly, journals a...
terraform init opens a new TCP connection for every provider registry request
While looking at why `terraform init` is slow, I found that it never reuses a connection to the provider registry. `RegistrySource` builds a new `http.Client`, with its own connection pool, for every request. So each provider's version lookup and each package metadata request...
CVE-2026-78662: Go x/crypto/ssh: prevent DoS on deadlocked undecided channel
...failing due to new CVE-2026-78662 ### What did you expect to happen? CVE scan should be clean ### How can we reproduce it (as minimally and precisely as possible)? run 'snyk test' against latest k/k master ### Anything else we need to know? _No response_ ### Kubernetes version...
`collect_all_async` blocks the event loop instead of returning almost instantly
### Checks - [x] I have checked that this issue has not already been reported. - [x] I have confirmed this bug exists on the [latest version](https://pypi.org/project/polars/) of Polars. ### Reproducible example ### Log output ### Issue description `pl.collect_all_async` runs...
Companies and products named
About this problem
Evidence
16 posts from 16 people in 12 places, about 2 a week over 58 days. Mostly on pola-rs/polars, mistralai/mistral-vibe, vllm-project/vllm. Tools named alongside: PyPI, Polars, Claude Code, Claude.
Frustration Frustration 1 of 3· Seen on GitHub issues, Hacker News, Reddit
How it was grouped
Posts that state a pain and match the "db-scaling" rule. First post Sep 2, 2026, latest Oct 29, 2026. Corroborated: 3 or more posts from 2 or more people or places. Method
The same problem elsewhere
GitHub hits connection, locking and scaling limits
6 posts from 6 people on GitHub issues.
MLflow hits connection, locking and scaling limits
4 posts from 4 people on GitHub issues.
PostgreSQL hits connection, locking and scaling limits
3 posts from 3 people on Reddit, GitHub issues, Stack Overflow.
PyTorch hits connection, locking and scaling limits
2 posts from 2 people on GitHub issues. Unverified.
ClickHouse hits connection, locking and scaling limits
2 posts from 2 people on GitHub issues. Unverified.
Mistral hits connection, locking and scaling limits
2 posts from 2 people on GitHub issues. Unverified.
Anthropic hits connection, locking and scaling limits
2 posts from 2 people on GitHub issues. Unverified.
History