Comments (4)
As I suspected, test_repair_mv
(and probably the other repair tests listed above) assumes that because they used RF=3 N=3, one can read from the view right after finishing to write to the base and be able to read everything. With tablets, this assumption is no longer correct, so the test sometimes, depending on timing, fails (it also sometimes succeeds, unsurprisingly).
The fix can to either enable synchronous_updates on the view in this test, or to add the missing "eventually".
It's easier for me to just add synchronous_updates - and it will also be less risky in the sense that I might miss one of the reads that needs an "eventually".
The downside of synchronous_updates is that will not work on Cassandra, but I don't think we ever cared and don't need to start caring now (these tests would have failed on Cassandra too because of the missing eventually).
UPDATE: The state of this test is even worse than I hoped. It didn't just assume that view updates are synchronous - even stranger, it assumed that it's possible to write with the default CL (1) and then read with the default CL (1). I don't know how this ever worked. Maybe it is more likely to work on overloaded test machines than on mine. I think I'll need to add eventually() instead of fixing every weirdness in this test separately.
UPDATE#2: Even with eventually, the test sometimes fails. I'm not sure why :-( Need to investigate further.
from scylladb.
Isn't synchronous view a different feature that needs to be tested, and 'eventually' is the 'default' mode we have and need to support?
from scylladb.
Isn't synchronous view a different feature that needs to be tested, and 'eventually' is the 'default' mode we have and need to support?
Yes, sort of. In theory, this test was supposed to be about repair, so could have used other features freely, and moreover the part of the test which fails is the setup - it's not even the real part of the test. But you're right that it's better to not depend on synchronous view updates if we don't have to. I won't be using it anyway, I switched to using eventually.
from scylladb.
The test_repair_mv
failure turns out to be cause by a known (but previously undocumented in an issue) implication of the disabling of self-pairing that we did with MV+tablets (commit 300e549). I created an issue to describe this implication and its possible solution in the future - #17043. But in the short term, the test can and should be fixed to not do CL=1 reads where they are not guaranteed to be enough. I'll send a fix for that shortly.
from scylladb.
Related Issues (20)
- docs: Issue on page Nodetool refresh HOT 2
- docs: Issue on page ScyllaDB Architecture - Fault Tolerance HOT 1
- Rewrite move_tablet API to go throw the topology coordinator as all other requests. HOT 2
- Failed to decommission second node due `raft operation [read_barrier] timed out`
- [tablets] when reducing the replication factor scylla should preferably drop replicas on dead nodes
- docs: add new branch 6.0
- [tablets] test_repair_streams_data_from_closest_node: AssertionError: New node should fetched data only from the same dc
- [tablets] test_remove_node_during_index_build: AssertionError: Expected a row count of 100000 in query "SELECT * FROM b_index_index", but got 99639
- service levels on raft: ugly error while creating service level in recovery mode HOT 4
- scylla_cpuset_setup file is off in ScyllaDB's AMI HOT 9
- Increasing RF doesn't copy mutations around properly HOT 6
- [RFE] Support Prometheus Service Discovery HOT 5
- service/migration_manager.cc:891:71: warning: 'view' used after it was moved [bugprone-use-after-move]
- Missing MV output in server-side DESCRIBE HOT 5
- [aarch64] Scylla dies on keyspace RF change in debug mode (asan) HOT 4
- [tablets] test_replace_node_*_ip_take_write dtests fail with: No mapping for :: in the passed effective replication map HOT 2
- [RFE] nodetool command to get tombstone age distribution
- validate the schema of uploaded sstables
- alternator: test_item_latency is flaky HOT 2
- raft topology: documentation: a node which failed to bootstrap (join the cluster) is permanently banned HOT 1
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 scylladb.