Adding presto-pinot with pushdown working #13504

agrawaldevesh · 2019-10-05T22:30:08Z

Adding presto-pinot with pushdown

This presto-pinot connector works as follows:

It builds upon the new plan connector optimizer framework to push the
maximal query to the pinot system.
By default, aggregation and limit queries are send to the pinot broker,
while just filter queries are send to the servers parallelly. This is
useful when joining against another (say hive) table wherein the join
needs to be parallelized and segments from pinot (fetched directly from
the pinot servers) need to be joined with hive splits. If you wish to
not send any queries to the broker, please set the catalog session
property prefer_broker_queries to false.
You can disable parallel queries (useful for low latency deployment)
we do at Uber, by setting the catalog property
forbid_segment_queries=false.
The connector needs a way to talk to the controller (to fetch the
brokers and the table schema). You can either provide the controller
URI or the URI of the Muttley proxy (an RPC proxy mechanism we use at
Uber). It is also possible to use the pinot-rest-proxy instead of
going to the pinot-broker (again something we use for our test pinot
clusters at Uber).

In addition the pinot connector would split the filter expressions such
that expressions that can be pushed down to pinot are done so, and those
that aren't are handled in presto.

It handles the full range of pinot PQL and respects pinot's weirdnesses
to guarantee that all results are still in the expected SQL semantics.

Usage: See the newly added connector/pinot.rst documentation file.

== RELEASE NOTES ==

Pinot Changes
* Added a new Pinot connector

SPI Changes
* Split `ConnectorPlanOptimizerProvider` to have two phases for connectors to participate query optimization: `LOGICAL` and `PHYSICAL`. The two phases correspond to pre- and post-shuffle optimization respectively.

highker

Will be unblocked by #13521

agrawaldevesh · 2019-10-19T02:55:46Z

@highker @wenleix

I think this PR is ready enough to start to be reviewed.

It depends on #13521. It also subsumes #13413 and has implemented all of the changes suggested in there. (cc: @haibow)

I know it is kind of big .. it has almost no changes to the presto-main except:

A way to run the ApplyConnectorOptimization pre exchange and tell it that it is being run pre exchange. Currently only the presto-pinot connector is implementing the pre-exchange optimizer. So it does not change the behavior for others that still run post exchanges.
It adds some getters for unit testing

The build will fail until https://github.com/agrawaldevesh/presto-pinot-driver can be pushed to mavencentral and the presto-pinot-driver version upgraded to 0.1.0. I am working with @wenleix on that.

I am also happy to meet in person and give a code overview. But a good place to start is from the PinotConnectorPlanOptimizer.java file.

highker

High-level comments only. The dependency looks much cleaner!

presto-pinot/presto-pinot-plugin/pom.xml

presto-pinot/presto-pinot-lib/pom.xml

presto-pinot/presto-pinot-lib/src/main/java/com/facebook/presto/pinot/PinotConnectorId.java

...ot/presto-pinot-lib/src/main/java/com/facebook/presto/pinot/PinotConnectorPlanOptimizer.java

presto-pinot/presto-pinot-plugin/src/main/java/com/facebook/presto/pinot/PinotPlugin.java

presto-pinot/presto-pinot-lib/src/main/java/com/facebook/presto/pinot/PinotConnector.java

...-pinot/presto-pinot-lib/src/main/java/com/facebook/presto/pinot/PinotPageSourceProvider.java

presto-pinot/presto-pinot-lib/src/main/java/com/facebook/presto/pinot/PinotSplit.java

presto-pinot/presto-pinot-lib/src/main/java/com/facebook/presto/pinot/PinotConfig.java

presto-pinot/presto-pinot-lib/src/main/java/com/facebook/presto/pinot/PinotConnection.java

...to-pinot/presto-pinot-lib/src/main/java/com/facebook/presto/pinot/PinotConnectorFactory.java

agrawaldevesh · 2019-10-22T14:14:47Z

We will then soon be having presto-aresdb-lib and presto-aresdb-plugin too :) we have other connector Pr’s coming up soon after this. I was just wary of polluting the top level directory too much :) But fine either which way.

…

On Tue, Oct 22, 2019 at 12:44 AM James Sun ***@***.***> wrote: ***@***.**** commented on this pull request. ------------------------------ In presto-pinot/presto-pinot-plugin/pom.xml <https://urldefense.proofpoint.com/v2/url?u=https-3A__github.com_prestodb_presto_pull_13504-23discussion-5Fr337366442&d=DwMCaQ&c=r2dcLCtU9q6n0vrtnDw9vg&r=Z3aBmCzO3AGueh3ZEPa3n4ujmbgsJJDE-x0U4W8t_Us&m=y3UguWObcP8n20TiTSq0sZydEIjOlYx2BQKQxUuud0k&s=uzXJQnJhEMdbGfHzCnpE1pkFaAxMoZQOpfPaSE4RbYk&e=> : > @@ -0,0 +1,63 @@ +<?xml version="1.0" encoding="UTF-8"?> Got it. It is similar to presto-hive and presto-hive-hadoop2. So instead of making the layout like presto-pinot/presto-pinot-lib and presto-pinot/presto-pinot-plugin, could you just make two top level modules? - presto-pinot-connector (this is the connector without plugin) - presto-pinot (this is the plugin) Although I don't quite like the naming, @wenleix <https://urldefense.proofpoint.com/v2/url?u=https-3A__github.com_wenleix&d=DwMCaQ&c=r2dcLCtU9q6n0vrtnDw9vg&r=Z3aBmCzO3AGueh3ZEPa3n4ujmbgsJJDE-x0U4W8t_Us&m=y3UguWObcP8n20TiTSq0sZydEIjOlYx2BQKQxUuud0k&s=Swugfa9sMMs3gmj_WUSxEJ7tBw_XORPSB7BdEkO-1FM&e=>, @arhimondr <https://urldefense.proofpoint.com/v2/url?u=https-3A__github.com_arhimondr&d=DwMCaQ&c=r2dcLCtU9q6n0vrtnDw9vg&r=Z3aBmCzO3AGueh3ZEPa3n4ujmbgsJJDE-x0U4W8t_Us&m=y3UguWObcP8n20TiTSq0sZydEIjOlYx2BQKQxUuud0k&s=f0UgQdv6ISJYY3iWbeMOu5QD7MLBqpVUeY4AcH0u4OM&e=> do you have better suggestions? — You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub <https://urldefense.proofpoint.com/v2/url?u=https-3A__github.com_prestodb_presto_pull_13504-3Femail-5Fsource-3Dnotifications-26email-5Ftoken-3DAE44R7ZWAEDIV7CRNUYVUCDQP2VMPA5CNFSM4I5ZSFXKYY3PNVWWK3TUL52HS4DFWFIHK3DMKJSXC5LFON2FEZLWNFSXPKTDN5WW2ZLOORPWSZGOCIXBMLY-23discussion-5Fr337366442&d=DwMCaQ&c=r2dcLCtU9q6n0vrtnDw9vg&r=Z3aBmCzO3AGueh3ZEPa3n4ujmbgsJJDE-x0U4W8t_Us&m=y3UguWObcP8n20TiTSq0sZydEIjOlYx2BQKQxUuud0k&s=9fQruJ7GyDWcXnT31_1UCwlgVeMQ2FGsY7lxoO66T-4&e=>, or unsubscribe <https://urldefense.proofpoint.com/v2/url?u=https-3A__github.com_notifications_unsubscribe-2Dauth_AE44R73V4CGDHGGUFYSXR3DQP2VMPANCNFSM4I5ZSFXA&d=DwMCaQ&c=r2dcLCtU9q6n0vrtnDw9vg&r=Z3aBmCzO3AGueh3ZEPa3n4ujmbgsJJDE-x0U4W8t_Us&m=y3UguWObcP8n20TiTSq0sZydEIjOlYx2BQKQxUuud0k&s=kfCkyDApfw59nv2vt1V8f1mx-NpW9QUIVC5B7QyOOVk&e=> .

highker · 2019-10-22T20:44:41Z

It is ok to have many top-level modules

sachdevs

Thanks for the contribution @agrawaldevesh! A couple of high level comments to get started:

presto-pinot-toolkit/src/main/java/com/facebook/presto/pinot/PinotBrokerPageSource.java

highker

Finished reviewing classes PinotExpression, PinotFilterExpressionConverter, PinotProjectExpressionConverter, and PinotQueryGenerator

presto-pinot-toolkit/pom.xml

presto-pinot-toolkit/src/main/java/com/facebook/presto/pinot/query/PinotExpression.java

...ot-toolkit/src/main/java/com/facebook/presto/pinot/query/PinotFilterExpressionConverter.java

presto-pinot-toolkit/src/main/java/com/facebook/presto/pinot/query/PinotQueryGenerator.java

...-pinot-toolkit/src/main/java/com/facebook/presto/pinot/query/PinotQueryGeneratorContext.java

presto-pinot-toolkit/src/main/java/com/facebook/presto/pinot/PinotConnectorPlanOptimizer.java

presto-pinot-toolkit/src/main/java/com/facebook/presto/pinot/PinotSegmentPageSource.java

presto-pinot-toolkit/src/main/java/com/facebook/presto/pinot/PinotSessionProperties.java

presto-pinot-toolkit/src/main/java/com/facebook/presto/pinot/PinotConfig.java

presto-pinot-toolkit/src/test/java/com/facebook/presto/pinot/MetadataUtil.java

presto-pinot-toolkit/src/main/java/com/facebook/presto/pinot/PushdownUtils.java

agrawaldevesh · 2019-10-23T22:28:17Z

@highker @sachdevs ... Thanks a lot for the detailed review !

I have incorporated your feedback and updated the code.

highker

@agrawaldevesh, there are still some amount of comments not addressed? Could you go through those in https://github.com/prestodb/presto/pull/13504/files ? Make sure to open each file and exam the comment.

presto-pinot-toolkit/src/main/java/com/facebook/presto/pinot/query/PinotExpression.java

...ot-toolkit/src/main/java/com/facebook/presto/pinot/query/PinotFilterExpressionConverter.java

highker

There are many nits and small coding style to be fixed. I only skimmed through the files rather than digging into each individual. @agrawaldevesh, could do a thorough exam on all the files to make sure the coding style is right based on the existing comments? Thanks!

...-pinot-toolkit/src/main/java/com/facebook/presto/pinot/query/PinotQueryGeneratorContext.java

presto-pinot-toolkit/src/main/java/com/facebook/presto/pinot/PinotBrokerPageSource.java

presto-pinot-toolkit/src/main/java/com/facebook/presto/pinot/PinotMetricsStat.java

presto-pinot-toolkit/src/main/java/com/facebook/presto/pinot/PinotSessionProperties.java

highker · 2019-10-24T06:27:25Z

presto-pinot-toolkit/src/main/java/com/facebook/presto/pinot/PushdownUtils.java

+import java.util.List;
+import java.util.Optional;
+
+public class PushdownUtils


Let's call this PinotPushdownUtils

Hmm ... something to discuss about: I want to use these for presto-aresdb connector too. We originally had them in presto-spi as a part of our fork. Is there a designated common module that connectors can share code in ? Or I can always have the presto-aresdb-toolkit depend on presto-pinot-toolkit but that would look weird.

It would be funny to create another module named presto-pinot-aresdb-utils to house these :)

But I can rename them to something pinot specific for now but this is generic.

presto-pinot-toolkit/src/main/java/com/facebook/presto/pinot/PushdownUtils.java

agrawaldevesh · 2019-10-24T19:07:15Z

@highker ... I just did a pass over the entire patch and I think you will find that a lot of your common nits have been addressed :-)

hellium01

Some initial comments.

hellium01 · 2019-10-24T17:04:49Z

...bc/src/main/java/com/facebook/presto/plugin/jdbc/optimization/JdbcPlanOptimizerProvider.java

-    public Set<ConnectorPlanOptimizer> getConnectorPlanOptimizers()
+    public Set<ConnectorPlanOptimizer> getLogicalPlanOptimizers()
+    {
+        return ImmutableSet.of();


Another way is to add a getStage() method in ConnectorPlanOptimizer but current design is fine given we won't (probably shouldn't) have more than two stages.

...t-toolkit/src/test/java/com/facebook/presto/pinot/query/TestPinotConnectorPlanOptimizer.java

hellium01 · 2019-10-24T17:29:21Z

...t-toolkit/src/test/java/com/facebook/presto/pinot/query/TestPinotConnectorPlanOptimizer.java

+     * A simplified table scan matcher for multiple-connector support.
+     * The goal is to test plan structural matching rather than table scan details
+     * <p>
+     * Copied from com.facebook.presto.sql.planner.optimizations.TestConnectorOptimization.SimpleTableScanMatcher


This comment might not be necessary?

...t-toolkit/src/test/java/com/facebook/presto/pinot/query/TestPinotConnectorPlanOptimizer.java

hellium01 · 2019-10-24T17:36:16Z

...t-toolkit/src/test/java/com/facebook/presto/pinot/query/TestPinotConnectorPlanOptimizer.java

+    }
+
+    @Test
+    public void testFilterSplitting()


testPartialPredicatePushdown?

...ot-toolkit/src/main/java/com/facebook/presto/pinot/query/PinotFilterExpressionConverter.java

presto-pinot-toolkit/src/main/java/com/facebook/presto/pinot/PinotErrorCode.java

hellium01 · 2019-10-24T18:55:07Z

...ot-toolkit/src/main/java/com/facebook/presto/pinot/query/PinotFilterExpressionConverter.java

+                return handleIn(specialForm, true, context);
+            case AND:
+            case OR:
+                return derived(format("(%s %s %s)",


Do we want to put parenthesis around sub clauses:

format("( (%s) %s (%s))", ...)

...ot-toolkit/src/main/java/com/facebook/presto/pinot/query/PinotFilterExpressionConverter.java

hellium01 · 2019-10-24T19:14:06Z

presto-pinot-toolkit/src/main/java/com/facebook/presto/pinot/query/PinotQueryGenerator.java

+            int aggregations = 0;
+            boolean groupByExists = false;
+
+            for (AggregationColumnNode expression : computeAggregationNodes(node).orElseThrow(() -> new PinotException(PINOT_UNSUPPORTED_EXPRESSION, Optional.empty(), "Unsupported aggregation node " + node))) {


nit: instead of lengthy computeAggregationNodes orElseThrow, can we just throw inside computeAggregationNode?

highker

The failure is related. Could you fix it really quick?

highker

@agrawaldevesh, could you update your commit message body with the following text? (Basically I just wrap your words into 70 chars per line and removed some out-of-date statement)

Adding presto-pinot with pushdown

This presto-pinot connector works as follows:
* It builds upon the new plan connector optimizer framework to push the
  maximal query to the pinot system.
* By default, aggregation and limit queries are send to the pinot broker,
  while just filter queries are send to the servers parallelly. This is
  useful when joining against another (say hive) table wherein the join
  needs to be parallelized and segments from pinot (fetched directly from 
  the pinot servers) need to be joined with hive splits. If you wish to 
  not send any queries to the broker, please set the catalog session 
  property prefer_broker_queries to false.
* You can disable parallel queries (useful for low latency deployment) 
  we do at Uber, by setting the catalog property
  forbid_segmnt_queries=false.
* The connector needs a way to talk to the controller (to fetch the 
  brokers and the table schema). You can either provide the controller 
  URI or the URI of the Muttley proxy (an RPC proxy mechanism we use at
  Uber). It is also possible to use the pinot-rest-proxy instead of 
  going to the pinot-broker (again something we use for our test pinot
  clusters at Uber).

In addition the pinot connector would split the filter expressions such
that expressions that can be pushed down to pinot are done so, and those
that aren't are handled in presto.

It handles the full range of pinot PQL and respects pinot's weirdnesses
to guarantee that all results are still in the expected SQL semantics.

Usage:
Place the catalog file like so in the etc/catalog directory. The schema
name is immaterial and it can be anything. (since pinot does not have a
notion of schema). Example catalog files can be found below:

$ cat local_run/etc/catalog/pinotnotparallel.properties
connector.name=pinot
pinot.controller-url=controller_host:9000
pinot.scan-parallelism-enabled=false

$ cat local_run/etc/catalog/pinotparallel.properties
connector.name=pinot
pinot.controller-url=controller_host:9000
pinot.scan-parallelism-enabled=true

highker

LGTM!

This presto-pinot connector works as follows: * It builds upon the new plan connector optimizer framework to push the maximal query to the pinot system. * By default, aggregation and limit queries are send to the pinot broker, while just filter queries are send to the servers parallelly. This is useful when joining against another (say hive) table wherein the join needs to be parallelized and segments from pinot (fetched directly from the pinot servers) need to be joined with hive splits. If you wish to not send any queries to the broker, please set the catalog session property prefer_broker_queries to false. * You can disable parallel queries (useful for low latency deployment) we do at Uber, by setting the catalog property forbid_segment_queries=false. * The connector needs a way to talk to the controller (to fetch the brokers and the table schema). You can either provide the controller URI or the URI of the Muttley proxy (an RPC proxy mechanism we use at Uber). It is also possible to use the pinot-rest-proxy instead of going to the pinot-broker (again something we use for our test pinot clusters at Uber). In addition the pinot connector would split the filter expressions such that expressions that can be pushed down to pinot are done so, and those that aren't are handled in presto. It handles the full range of pinot PQL and respects pinot's weirdnesses to guarantee that all results are still in the expected SQL semantics. Usage: See the newly added connector/pinot.rst documentation file.

agrawaldevesh · 2019-10-26T05:39:33Z

Tested and fixed. Good to be merged in.

…o's date_trunc This will only be called by Presto connector (prestodb/presto#13504) and it has identically semantics to presto's SQL date_trunc, albeit with a timezone specialization. Please see details on presto's date_trunc function here: https://prestodb.github.io/docs/current/functions/datetime.html This is needed so that the presto's date_trunc invocations can be faithfully translated as is to this new function. Its a new function so that it is trivial to roll out to harmlessly without a lot of regression testing. Without this function, we cannot handle timezones nor week truncations in the existing pinot's dateTimeConvert function. It basically copies the PrestoDB code: https://github.com/prestodb/presto/blob/master/presto-main/src/main/java/com/facebook/presto/operator/scalar/DateTimeFunctions.java, (specializations of date_trunc for TIMESTAMP and TIMESTAMP_WITH_TIME_ZONE I am even checking in the zone-index.properties used by presto to ensure that even the time zones are 1:1 b/w presto and this function. (sync'd to the latest prestodb/presto repo) Understanding this UDF requires knowledge of the joda-time API. I am not documenting this heavily since it is a copy of the Presto UDF.

…o's date_trunc (#4740) This will only be called by Presto connector (prestodb/presto#13504) and it has identically semantics to presto's SQL date_trunc, albeit with a timezone specialization. Please see details on presto's date_trunc function here: https://prestodb.github.io/docs/current/functions/datetime.html This is needed so that the presto's date_trunc invocations can be faithfully translated as is to this new function. Its a new function so that it is trivial to roll out to harmlessly without a lot of regression testing. Without this function, we cannot handle timezones nor week truncations in the existing pinot's dateTimeConvert function. It basically copies the PrestoDB code: https://github.com/prestodb/presto/blob/master/presto-main/src/main/java/com/facebook/presto/operator/scalar/DateTimeFunctions.java, (specializations of date_trunc for TIMESTAMP and TIMESTAMP_WITH_TIME_ZONE I am even checking in the zone-index.properties used by presto to ensure that even the time zones are 1:1 b/w presto and this function. (sync'd to the latest prestodb/presto repo) Understanding this UDF requires knowledge of the joda-time API. I am not documenting this heavily since it is a copy of the Presto UDF.

…o's date_trunc (apache#4740) This will only be called by Presto connector (prestodb/presto#13504) and it has identically semantics to presto's SQL date_trunc, albeit with a timezone specialization. Please see details on presto's date_trunc function here: https://prestodb.github.io/docs/current/functions/datetime.html This is needed so that the presto's date_trunc invocations can be faithfully translated as is to this new function. Its a new function so that it is trivial to roll out to harmlessly without a lot of regression testing. Without this function, we cannot handle timezones nor week truncations in the existing pinot's dateTimeConvert function. It basically copies the PrestoDB code: https://github.com/prestodb/presto/blob/master/presto-main/src/main/java/com/facebook/presto/operator/scalar/DateTimeFunctions.java, (specializations of date_trunc for TIMESTAMP and TIMESTAMP_WITH_TIME_ZONE I am even checking in the zone-index.properties used by presto to ensure that even the time zones are 1:1 b/w presto and this function. (sync'd to the latest prestodb/presto repo) Understanding this UDF requires knowledge of the joda-time API. I am not documenting this heavily since it is a copy of the Presto UDF.

facebook-github-bot added the CLA Signed label Oct 5, 2019

highker self-requested a review October 6, 2019 05:33

agrawaldevesh force-pushed the devesh-pinot-test branch 2 times, most recently from 9ebfaa7 to ae9bb18 Compare October 9, 2019 06:30

highker reviewed Oct 9, 2019

View reviewed changes

agrawaldevesh force-pushed the devesh-pinot-test branch 3 times, most recently from ab074c2 to 563b1a5 Compare October 19, 2019 02:43

highker changed the title ~~[WIP DO NOT REVIEW] Adding presto-pinot with pushdown working~~ Adding presto-pinot with pushdown working Oct 19, 2019

highker requested review from wenleix and highker October 19, 2019 03:20

highker mentioned this pull request Oct 19, 2019

Add Presto Pinot Connector #13413

Closed

sachdevs self-requested a review October 19, 2019 17:44

highker requested a review from hellium01 October 20, 2019 06:17

highker reviewed Oct 20, 2019

View reviewed changes

hellium01 reviewed Oct 22, 2019

View reviewed changes

agrawaldevesh mentioned this pull request Oct 22, 2019

Minimal presto-pinot driver exposing just pinot-common and some helper prestodb/presto-pinot-driver#1

Merged

agrawaldevesh force-pushed the devesh-pinot-test branch from 563b1a5 to de579bf Compare October 22, 2019 06:27

highker requested a review from zhenxiao October 22, 2019 22:45

agrawaldevesh force-pushed the devesh-pinot-test branch from de579bf to 0bca794 Compare October 22, 2019 23:14

highker self-requested a review October 23, 2019 01:40

sachdevs reviewed Oct 23, 2019

View reviewed changes

agrawaldevesh force-pushed the devesh-pinot-test branch from 0bca794 to 284b4de Compare October 23, 2019 05:23

highker reviewed Oct 23, 2019

View reviewed changes

sachdevs reviewed Oct 23, 2019

View reviewed changes

agrawaldevesh force-pushed the devesh-pinot-test branch from 284b4de to 34e8204 Compare October 23, 2019 22:27

highker reviewed Oct 24, 2019

View reviewed changes

agrawaldevesh force-pushed the devesh-pinot-test branch from 9661d49 to 1830602 Compare October 24, 2019 18:58

hellium01 reviewed Oct 24, 2019

View reviewed changes

highker self-requested a review October 24, 2019 23:20

highker reviewed Oct 25, 2019

View reviewed changes

agrawaldevesh force-pushed the devesh-pinot-test branch from 1830602 to e8b3174 Compare October 25, 2019 02:44

highker approved these changes Oct 25, 2019

View reviewed changes

highker mentioned this pull request Oct 25, 2019

[Design] Connector Participating Query Optimization #13102

Closed

3 tasks

agrawaldevesh force-pushed the devesh-pinot-test branch from e8b3174 to af6bb6a Compare October 26, 2019 03:22

agrawaldevesh force-pushed the devesh-pinot-test branch from af6bb6a to 95b935b Compare October 26, 2019 04:11

highker merged commit 78f2470 into prestodb:master Oct 26, 2019

aaneja mentioned this pull request Jul 10, 2023

PredicatePushdown rule adds un-needed ProjectNode's #19749

Closed

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Adding presto-pinot with pushdown working #13504

Adding presto-pinot with pushdown working #13504

agrawaldevesh commented Oct 5, 2019 •

edited by highker

highker left a comment

agrawaldevesh commented Oct 19, 2019 •

edited

highker left a comment

agrawaldevesh commented Oct 22, 2019 via email

highker commented Oct 22, 2019

sachdevs left a comment

highker left a comment

agrawaldevesh commented Oct 23, 2019

highker left a comment

highker left a comment

highker Oct 24, 2019

agrawaldevesh Oct 24, 2019

agrawaldevesh commented Oct 24, 2019

hellium01 left a comment

hellium01 Oct 24, 2019

hellium01 Oct 24, 2019

hellium01 Oct 24, 2019

hellium01 Oct 24, 2019

hellium01 Oct 24, 2019

highker left a comment

highker left a comment

highker left a comment

agrawaldevesh commented Oct 26, 2019

Adding presto-pinot with pushdown working #13504

Adding presto-pinot with pushdown working #13504

Conversation

agrawaldevesh commented Oct 5, 2019 • edited by highker

highker left a comment

Choose a reason for hiding this comment

agrawaldevesh commented Oct 19, 2019 • edited

highker left a comment

Choose a reason for hiding this comment

agrawaldevesh commented Oct 22, 2019 via email

highker commented Oct 22, 2019

sachdevs left a comment

Choose a reason for hiding this comment

highker left a comment

Choose a reason for hiding this comment

agrawaldevesh commented Oct 23, 2019

highker left a comment

Choose a reason for hiding this comment

highker left a comment

Choose a reason for hiding this comment

highker Oct 24, 2019

Choose a reason for hiding this comment

agrawaldevesh Oct 24, 2019

Choose a reason for hiding this comment

agrawaldevesh commented Oct 24, 2019

hellium01 left a comment

Choose a reason for hiding this comment

hellium01 Oct 24, 2019

Choose a reason for hiding this comment

hellium01 Oct 24, 2019

Choose a reason for hiding this comment

hellium01 Oct 24, 2019

Choose a reason for hiding this comment

hellium01 Oct 24, 2019

Choose a reason for hiding this comment

hellium01 Oct 24, 2019

Choose a reason for hiding this comment

highker left a comment

Choose a reason for hiding this comment

highker left a comment

Choose a reason for hiding this comment

highker left a comment

Choose a reason for hiding this comment

agrawaldevesh commented Oct 26, 2019

agrawaldevesh commented Oct 5, 2019 •

edited by highker

agrawaldevesh commented Oct 19, 2019 •

edited