TEZ-3331: Add operation specific HDFS counters for Tez UI #379
Conversation
This comment was marked as outdated.
This comment was marked as outdated.
This comment was marked as outdated.
This comment was marked as outdated.
|
@ayushtkn, can you please review this one? we're using this downstream for a long time, haven't ported back to upstream yet |
| FILE_BYTES_READ("fileBytesRead"), | ||
| FILE_BYTES_WRITTEN("fileBytesWritten"), | ||
|
|
||
| // Additional counters from HADOOP-13305 |
There was a problem hiding this comment.
There seems to be some more counters as well, like OP_CREATE_NON_RECURSIVE, OP_EXISTS, OP_IS_FILE etc
Should we include them as well?
There was a problem hiding this comment.
yeah, I'm refreshing it according to the current state of CommonStatisticNames
| if (!statisticUpdaters.containsKey(stats.getScheme())) { | ||
| Map<String, FileSystemStatisticUpdater> updaterSet = new TreeMap<>(); | ||
| statisticUpdaters.put(stats.getScheme(), updaterSet); | ||
| } | ||
| FileSystemStatisticUpdater updater = statisticUpdaters.get(stats.getScheme()) | ||
| .computeIfAbsent(stats.getName(), k -> new FileSystemStatisticUpdater(tezCounters, stats)); |
There was a problem hiding this comment.
this looks like computeIfAbsent
maybe can do
// Fetch or initialize the updater set for the scheme
Map<String, FileSystemStatisticUpdater> updaterSet = statisticUpdaters
.computeIfAbsent(stats.getScheme(), k -> new TreeMap<>());
// Fetch or create the updater for the specific statistic
FileSystemStatisticUpdater updater = updaterSet
.computeIfAbsent(stats.getName(), k -> new FileSystemStatisticUpdater(tezCounters, stats));
There was a problem hiding this comment.
absolutely, thanks!
|
|
||
| private static MiniDFSCluster dfsCluster; | ||
|
|
||
| private static Configuration conf = new Configuration(); |
| public static void setup() throws IOException { | ||
| try { | ||
| conf.set(MiniDFSCluster.HDFS_MINIDFS_BASEDIR, TEST_ROOT_DIR); | ||
| dfsCluster = new MiniDFSCluster.Builder(conf).numDataNodes(2).format(true).racks(null) |
There was a problem hiding this comment.
.format(true).racks(null) ain't required IMO, they are by default true & null
| FSDataOutputStream out = remoteFs.create(new Path("/tmp/foo/abc.txt")); | ||
| out.writeBytes("xyz"); | ||
| out.close(); |
There was a problem hiding this comment.
can use this
DFSTestUtil.writeFile(remoteFs, new Path("/tmp/foo/abc.txt"), "xyz");
and below as well
There was a problem hiding this comment.
I can see there is a difference between how these approaches work, which is reflected in the counters
no matter which one we choose, the counters can be asserted accordingly
original approach (mkdir, create)
HDFS_BYTES_WRITTEN=3
HDFS_WRITE_OPS=2
HDFS_OP_CREATE=1
HDFS_OP_MKDIRS=1
DFSTestUtil.writeFile
HDFS_BYTES_WRITTEN=3
HDFS_READ_OPS=1
HDFS_WRITE_OPS=1
HDFS_OP_CREATE=1
HDFS_OP_GET_FILE_STATUS=1
|
|
||
| private static final Logger LOG = LoggerFactory.getLogger( | ||
| TestTaskCounterUpdater.class); | ||
| private static Configuration conf = new Configuration(); |
| private static final Logger LOG = LoggerFactory.getLogger( | ||
| TestTaskCounterUpdater.class); |
There was a problem hiding this comment.
nit
line break wasn't required I think
| TaskCounterUpdater updater = new TaskCounterUpdater(counters, conf, "pid"); | ||
|
|
||
| updater.updateCounters(); | ||
| LOG.info("Counters: " + counters); |
| Assert.assertTrue("Counter not updated, old=" + oldVal | ||
| + ", new=" + cpuCounter.getValue(), cpuCounter.getValue() > oldVal); |
There was a problem hiding this comment.
should we put some sleep before updateCounters, just thinking in extreme conditions, this check won't go flaky, right?
There was a problem hiding this comment.
not sure, it seems quite unlikely, I would let it be decided by the future's precommit runs (it will be obvious once it fails)
|
|
||
|
|
| OP_SET_ACL(CommonStatisticNames.OP_SET_ACL), | ||
| OP_SET_OWNER(CommonStatisticNames.OP_SET_OWNER), | ||
| OP_SET_PERMISSION(CommonStatisticNames.OP_SET_PERMISSION), | ||
| OP_GET_FILE_BLOCK_LOCATIONS("op_get_file_block_locations"); |
There was a problem hiding this comment.
what is this, no enum in Hadoop? where is this coming from then?
There was a problem hiding this comment.
it's from DFSOpsCountStatistics, but I'm removing it as I intend to include and expose only stats that are defined in CommonStatisticNames
|
thanks @ayushtkn for the initial comments! I need to rework this area, I think some of the comments are related to the fact that I simply adapted an old patch, I need to think this over again! I'll let you know |
abstractdog
force-pushed
the
TEZ-3331
branch
from
60db8f1 to
532adec
Compare
This comment was marked as outdated.
This comment was marked as outdated.
abstractdog
force-pushed
the
TEZ-3331
branch
from
532adec to
d642287
Compare
|
🎊 +1 overall
This message was automatically generated. |
There was a problem hiding this comment.
Thanx @abstractdog for the changes, Minor Comments, rest looks good
| // Additional counters from HADOOP-13305 | ||
| // Additional counters from HADOOP-13305 |
| // /** | ||
| // * A Map where Key-> URIScheme and value->FileSystemStatisticUpdater | ||
| // */ | ||
| // private Map<String, FileSystemStatisticUpdater> statisticUpdaters = | ||
| // new HashMap<>(); |
There was a problem hiding this comment.
why are you commenting this out? Can't we just delete it
| private Map<String, FileSystemStatisticUpdater> statisticUpdaters = | ||
| new HashMap<String, FileSystemStatisticUpdater>(); | ||
| private Map<String, Map<String, FileSystemStatisticUpdater>> statisticUpdaters = | ||
| new HashMap<>(); |
| StorageStatistics stats = iter.next(); | ||
| // Fetch or initialize the updater set for the scheme | ||
| Map<String, FileSystemStatisticUpdater> updaterSet = statisticUpdaters | ||
| .computeIfAbsent(stats.getScheme(), k -> new TreeMap<>()); |
There was a problem hiding this comment.
Why are we using TreeMap now? If I decode right, earlier it was HashMap, it would be some cost using it right?
|
|
||
| DFSTestUtil.writeFile(remoteFs, new Path("/tmp/foo/abc.txt"), "xyz"); | ||
|
|
||
| updater.updateCounters(); |
There was a problem hiding this comment.
shouldn't we first do
FileSystem.clearStatistics();
In case there is any test added in future which does FS operations, I believe this test will screw up. So better to reset everything to 0, before we start testing
This a rebased, reworked version of the last patch on TEZ-3331:
https://issues.apache.org/jira/secure/attachment/12926702/TEZ-3331.8.patch