[
https://issues.apache.org/jira/browse/CASSANDRA-15907?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17159912#comment-17159912
]
Andres de la Peña commented on CASSANDRA-15907:
-----------------------------------------------
[~maedhroz] I like the changes to the lazy rows approach. However, I'm afraid
we need the snapshot of the cached rows done in [this local
copy-and-clear|https://github.com/adelapena/cassandra/blob/accf2a47c341875942b0d8b06c016cc0d66d62cb/src/java/org/apache/cassandra/service/ReplicaFilteringProtection.java#L522-L524],
otherwise the advances in the other replica can introduce new data and mess
with it, producing multiple test failures. Or, we can do much better than in my
previous patch and just track the number of contents in the snapshot and save
us the queue copy, as it's done
[here|https://github.com/adelapena/cassandra/blob/35d8e712bbbe03076ba867c11759664e8ff839e4/src/java/org/apache/cassandra/service/ReplicaFilteringProtection.java#L528-L568].
Also I think that making {{currentMergedRows}} / {{unprotectedPartition}} to
the partition iterator is not correct. It's a pointer to the current first
iteration merged partition and it should be shared by all the builders in the
RFP. If we make it local it can reduce the speed at which that pointer is
advanced, producing in the end more RFP queries.
{quote}When there is a large number of non-conflicting rows at the start of the
first-phase iterator, though, it seems like the price of avoiding row caching
is creating a large number of {{CachedRowIterator}} objects. Maybe this is the
right trade-off, but I'm not sure.
{quote}
We can find a balance between max cache size and the number of
{{CachedRowIterator}} instances if we try to grow the cache a bit further when
there are no conflicts:
{code:java}
while (unprotectedPartition != null && unprotectedPartition.hasNext()
&& (toFetch != null || cachedRows.size() < min_cache_size))
{code}
Min cache/buffer size can be a constant, or a config property, or a function of
the warning threshold, or something related the query limit. This would still
limit the size of the cache in the absence of conflicts while quickly reducing
the number of {{CachedRowIterator}} instances.
Also, given that we are concerned about the cache size, we might want to
consider tracking the max size that the cache reaches during the query, and add
it to a new table metric that tracks the average max cache size.
> Operational Improvements & Hardening for Replica Filtering Protection
> ---------------------------------------------------------------------
>
> Key: CASSANDRA-15907
> URL: https://issues.apache.org/jira/browse/CASSANDRA-15907
> Project: Cassandra
> Issue Type: Improvement
> Components: Consistency/Coordination, Feature/2i Index
> Reporter: Caleb Rackliffe
> Assignee: Caleb Rackliffe
> Priority: Normal
> Labels: 2i, memory
> Fix For: 3.0.x, 3.11.x, 4.0-beta
>
> Time Spent: 4h 20m
> Remaining Estimate: 0h
>
> CASSANDRA-8272 uses additional space on the heap to ensure correctness for 2i
> and filtering queries at consistency levels above ONE/LOCAL_ONE. There are a
> few things we should follow up on, however, to make life a bit easier for
> operators and generally de-risk usage:
> (Note: Line numbers are based on {{trunk}} as of
> {{3cfe3c9f0dcf8ca8b25ad111800a21725bf152cb}}.)
> *Minor Optimizations*
> * {{ReplicaFilteringProtection:114}} - Given we size them up-front, we may be
> able to use simple arrays instead of lists for {{rowsToFetch}} and
> {{originalPartitions}}. Alternatively (or also), we may be able to null out
> references in these two collections more aggressively. (ex. Using
> {{ArrayList#set()}} instead of {{get()}} in {{queryProtectedPartitions()}},
> assuming we pass {{toFetch}} as an argument to {{querySourceOnKey()}}.)
> * {{ReplicaFilteringProtection:323}} - We may be able to use
> {{EncodingStats.merge()}} and remove the custom {{stats()}} method.
> * {{DataResolver:111 & 228}} - Cache an instance of
> {{UnaryOperator#identity()}} instead of creating one on the fly.
> * {{ReplicaFilteringProtection:217}} - We may be able to scatter/gather
> rather than serially querying every row that needs to be completed. This
> isn't a clear win perhaps, given it targets the latency of single queries and
> adds some complexity. (Certainly a decent candidate to kick even out of this
> issue.)
> *Documentation and Intelligibility*
> * There are a few places (CHANGES.txt, tracing output in
> {{ReplicaFilteringProtection}}, etc.) where we mention "replica-side
> filtering protection" (which makes it seem like the coordinator doesn't
> filter) rather than "replica filtering protection" (which sounds more like
> what we actually do, which is protect ourselves against incorrect replica
> filtering results). It's a minor fix, but would avoid confusion.
> * The method call chain in {{DataResolver}} might be a bit simpler if we put
> the {{repairedDataTracker}} in {{ResolveContext}}.
> *Testing*
> * I want to bite the bullet and get some basic tests for RFP (including any
> guardrails we might add here) onto the in-JVM dtest framework.
> *Guardrails*
> * As it stands, we don't have a way to enforce an upper bound on the memory
> usage of {{ReplicaFilteringProtection}} which caches row responses from the
> first round of requests. (Remember, these are later used to merged with the
> second round of results to complete the data for filtering.) Operators will
> likely need a way to protect themselves, i.e. simply fail queries if they hit
> a particular threshold rather than GC nodes into oblivion. (Having control
> over limits and page sizes doesn't quite get us there, because stale results
> _expand_ the number of incomplete results we must cache.) The fun question is
> how we do this, with the primary axes being scope (per-query, global, etc.)
> and granularity (per-partition, per-row, per-cell, actual heap usage, etc.).
> My starting disposition on the right trade-off between
> performance/complexity and accuracy is having something along the lines of
> cached rows per query. Prior art suggests this probably makes sense alongside
> things like {{tombstone_failure_threshold}} in {{cassandra.yaml}}.
--
This message was sent by Atlassian Jira
(v8.3.4#803005)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
