impala-reviews mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Bharath Vissapragada (Code Review)" <>
Subject [Impala-ASF-CR] IMPALA-5429: Multi threaded block metadata loading
Date Fri, 20 Oct 2017 21:09:38 GMT
Bharath Vissapragada has posted comments on this change. (

Change subject: IMPALA-5429: Multi threaded block metadata loading

Patch Set 6:

File be/src/catalog/
PS6, Line 38: DEFINE_int32(max_hdfs_parts_parallel_load, 5,
> Let's spell out "parts" into "partitions" to avoid confusion.
PS6, Line 40:     "tables. Due to HDFS architectural limitations, it is unlikely to get a
linear "
> In the sentence "Due to HDFS architectural..." what does the "it" after the
PS6, Line 42: DEFINE_int32(max_s3_parts_parallel_load, 20,
> Does ADLS fall into this or the HDFS bucket? What about other filesystems?
Yes, although I haven't tested it specifically with ADLS. I mentioned that in the commit message
but it looks like a good thing to reflect that in the config name too. I changed it too, let
me know if you think we should use a better name here.
File fe/src/main/java/org/apache/impala/catalog/
PS6, Line 216:   private class PathStorageMdLoadingStats {
> Naming seems a little weird. I think "Storage" is too generic and "Block" i
Wfm. Changed it.
PS6, Line 240:     // If set to true, reloads the block metadata only when the underlying
> Please clarify "the underlying file"
PS6, Line 413:         ++loadStats.ignoredFiles;
> I think we should distinguish the hidden file and already-up-to-date file c
PS6, Line 773:    * Returns the thread pool size to load the block metadata of this table.
> Command on how we distinguish between HDFS, S3 and other filesystems (using
PS6, Line 793:     return Math.max(Math.min(numPaths, threadPoolSize), 1);
> For my understanding, why is the max() needed here? Is it not be a precondi
For empty tables, numPaths is 0. But yes this should be a preconditions check. So I moved
it to the caller and added a preconditions check in this method.
PS6, Line 800:    * metadata is reloaded, else it is loaded from scratch.
> Difference between "reloaded" and "loaded from scratch" is not very clear. 
Sounds better. Done.
PS6, Line 804:     int numPathsToLoad = partsByPath.keySet().size();
> partsByPath.size()?
PS6, Line 812:       for (Path p: partsByPath.keySet()) {
> The tasks are ordered randomly. I wonder if submitting the tasks in a clust
Will do. I have two other jiras (based on others' comments) to create, so will add this to
the list.
PS6, Line 821:         } catch (ExecutionException | InterruptedException e) {
> We still consider the load successful if one of these fails. What do you th
Oops, I forgot to wrap it as a TLE and throw. Thanks for pointing it out. My opinion is that
a partially loaded table is ok, but we should let the user know about it. What do you think?
PS6, Line 822:           LOG.error("Encountered an error loading block metadata for table:
" +
> Let's also dump the path
ExecutionException (ex, IOException in this case) includes the path by default and LOG.error()
logs the full trace?
PS6, Line 898:    * the newly created HdfsPartitions in parallel.
> remove "the"
File fe/src/main/java/org/apache/impala/util/
PS6, Line 38:   private List<T> list_ = Collections.synchronizedList(new ArrayList<T>());
> The CC requirements and behavior are not clear to me. Why is it not suffici
My understanding is that it could affect the throughput of getIndex(). Once the cache is warmed
up with all the host mapping, we don't need it to be synchronized anymore? Is my understanding

To view, visit
To unsubscribe, visit

Gerrit-Project: Impala-ASF
Gerrit-Branch: master
Gerrit-MessageType: comment
Gerrit-Change-Id: I07eaa7151dfc4d56da8db8c2654bd65d8f808481
Gerrit-Change-Number: 8235
Gerrit-PatchSet: 6
Gerrit-Owner: Bharath Vissapragada <>
Gerrit-Reviewer: Alex Behm <>
Gerrit-Reviewer: Bharath Vissapragada <>
Gerrit-Reviewer: Dimitris Tsirogiannis <>
Gerrit-Reviewer: Jim Apple <>
Gerrit-Reviewer: Mostafa Mokhtar <>
Gerrit-Reviewer: Vuk Ercegovac <>
Gerrit-Comment-Date: Fri, 20 Oct 2017 21:09:38 +0000
Gerrit-HasComments: Yes

  • Unnamed multipart/alternative (inline, 8-Bit, 0 bytes)
View raw message