Comments (4)
As far as I know, the search core does not set or modify the ownership of its data directories or files. However—is this by any chance running on an XFS filesystem? We have observed this exact scenario very spuriously on XFS during periods with lots of IO and file syncing, but only when the kernel file system buffer cache and/or directory entries cache is being concurrently flushed/evicted.
The observed result was a directory with partially bogus inode metadata, in particular an owner that does not make sense (we've seen uid/gid of both 65534 and 0, neither of which are correct). Also note the Links: 1
field—I would expect a correct directory inode to have at least 2 links since a directory should normally include a hard-link to itself (note; does not apply to all file systems—Btrfs for instance seems to just have 1).
The scenario you describe is one where the kernel is under a very high memory pressure, and where it's likely to try to evict file system caches to get more available memory for applications. If so, the preconditions are similar to what we have observed.
Note: this is inherently just a theory, and we have not had the time to do a deep dive into figuring this out (particularly since doing so would involve reading lots of kernel code). I've tried reproducing this issue by writing a program that stresses the kernel file system cache management through lots of concurrent nested directory creation, fsyncing and forced cache eviction, but never managed to trigger it.
from vespa.
It is XFS indeed
from vespa.
Entirely anecdotally, we have not observed this issue for some time on our systems. We do regular fleet-wide OS upgrades, so it's possible the underlying issue has been fixed in the upstream kernel.
What's the kernel version you are running? I.e. uname -rv
.
For instance, torvalds/linux@28b4b05 (which I have no idea if is related, but does sound strikingly familiar...) appears to be part of kernel v6.2 and beyond. After doing some digging it looks like this particular fix was backported into RHEL8 kernel version 4.18.0-513.5.1
.
from vespa.
Related Issues (20)
- Support setting metrics-proxy heap size HOT 3
- All Search Nodes are crashing HOT 4
- Potential Memory Leak Issue in PrometheusModel class HOT 4
- Node coming up with latest deployed application version after being down HOT 5
- Configurable max token length HOT 1
- Add embedding instruction prompt support for to hf-embedder HOT 1
- NPE while deploying HOT 2
- Cluster Crashes When Distribution Key Is Too High HOT 3
- Error code - ORT_INVALID_PROTOBUF HOT 2
- Add deploy time warning about combining paged attributes with index (hnsw) or fast-search
- you probably have an older CPU than required HOT 7
- Unable to install Vespa version 8.334.22, dependencies not found HOT 2
- Tokenizer.json compability with ai.djl.huggingface tokenizers broken for sentencepiece based models HOT 5
- Add response degradation indicator for approximate nearest neighbor
- Support userQuery/userInput in the JSON query format
- Sort ASC by a field should give those results without that field value HOT 7
- Please provide a simple UI interface to applications like colbert, colbert-long, etc HOT 2
- baseport for content cluster nodes doesn't work well for all node types
- Skewed distribution of buckets across content cluster groups HOT 6
Recommend Projects
-
React
A declarative, efficient, and flexible JavaScript library for building user interfaces.
-
Vue.js
🖖 Vue.js is a progressive, incrementally-adoptable JavaScript framework for building UI on the web.
-
Typescript
TypeScript is a superset of JavaScript that compiles to clean JavaScript output.
-
TensorFlow
An Open Source Machine Learning Framework for Everyone
-
Django
The Web framework for perfectionists with deadlines.
-
Laravel
A PHP framework for web artisans
-
D3
Bring data to life with SVG, Canvas and HTML. 📊📈🎉
-
Recommend Topics
-
javascript
JavaScript (JS) is a lightweight interpreted programming language with first-class functions.
-
web
Some thing interesting about web. New door for the world.
-
server
A server is a program made to process requests and deliver data to clients.
-
Machine learning
Machine learning is a way of modeling and interpreting data that allows a piece of software to respond intelligently.
-
Visualization
Some thing interesting about visualization, use data art
-
Game
Some thing interesting about game, make everyone happy.
Recommend Org
-
Facebook
We are working to build community through open source technology. NB: members must have two-factor auth.
-
Microsoft
Open source projects and samples from Microsoft.
-
Google
Google ❤️ Open Source for everyone.
-
Alibaba
Alibaba Open Source for everyone
-
D3
Data-Driven Documents codes.
-
Tencent
China tencent open source team.
from vespa.