With plugins of a meta-plugin now loaded in separate classloaders, we
should not be loading all classes in all plugins of a meta-plugin when
executing scripts. This is particularly problematic in the case of
security extensions where the install plugin extension command would be
running with the classpath of all plugins. However, if there is JAR hell
in this classpath, installation would fail. This is not realistic though
since the plugins are run in separate classloaders. To fix this, for the
scripts of a plugin, we only set the classpath to include the JARs for
that plugin and the JARs of core. This leads us to the introduction of
plugin-specific env scripts.
Relates elastic/x-pack-elasticsearch#3649
Original commit: elastic/x-pack-elasticsearch@543df37eed
Some imports were changed in 6.x to address line-length issues
there. This commit pulls the same changes to master to keep the branches
consistent to simplify backports.
Original commit: elastic/x-pack-elasticsearch@190f9d41f5
This is no longer needed for general X-Pack since the tests will not be
depending on ML. We move this class to an ML specific directory and
remove the dependency from other tests.
Original commit: elastic/x-pack-elasticsearch@9b287f7460
This allows CSV tests to include a line between the header and the
values that is ignored by the test framework. This optional line can be
added to the tests to make them a little easier to read which is useful
when they are included in the documentation. As a side effect they also
closely mimick the output of the CLI. To the point where you can copy
directly from the CLI and paste into the CSV tests.
Example:
```
constantYear
// tag::year
SELECT YEAR(CAST('2018-01-19T10:23:27Z' AS TIMESTAMP)) as year;
year
---------------
2018
// end::year
;
```
This can be extracted with a construct like this in the docs:
```
["source","sql",subs="attributes,callouts,macros"]
--------------------------------------------------
include-tagged::{sql-specs}/datetime.csv-spec[year]
--------------------------------------------------
```
Which makes documentation that looks like this:
```
SELECT YEAR(CAST('2018-01-19T10:23:27Z' AS TIMESTAMP)) as year;
year
---------------
2018
```
Which is fairly nice.
Original commit: elastic/x-pack-elasticsearch@8c10b5cb10
I went to write some docs for datetime functions that look like:
```
SELECT YEAR(CAST('2018-01-19T10:23:27Z' AS TIMESTAMP)) as year;
year
2018
```
because I figured they'd be pretty easy to read because they didn't
require any knowledge of a data set. But it turns out that constant
folding doesn't work properly for date time functions because they don't
actually apply the extraction.
Original commit: elastic/x-pack-elasticsearch@aa9c66b2c7
This commit splits the transport implementations into components that
can be used client-side (in the transport client) and server-side (in
the server). This enables removing security as a dependency for the
transport client.
Relates elastic/x-pack-elasticsearch#3635
Original commit: elastic/x-pack-elasticsearch@e480eb7eb2
Since we don't do any serialization of Cursor classes from the Transport
side (it is treated entirely as a string), we don't actually need to tell anyone
about SQL's writeables
Original commit: elastic/x-pack-elasticsearch@ad2c10e327
Rather than run every SQL test against multiple nodes we instead run a
subset of the tests simply for speed. The subset is both small but
exercises the CLI, JDBC, and REST interface and a few special "these
might be different in against multiple nodes" cases.
This drops the run time from of `gradle clean check` of these tests from
```
BUILD SUCCESSFUL in 4m 49s
```
To:
```
BUILD SUCCESSFUL in 2m 49s
```
Original commit: elastic/x-pack-elasticsearch@dcd99bcd82
This is related to elastic/elasticsearch#elastic/x-pack-elasticsearch#28275. It modifies x-pack to
support the changes in channel contexts. Additionally, it simplifies
the SSLChannelContext by relying on some common work between it and
BytesChannelContext.
Original commit: elastic/x-pack-elasticsearch@8a8fcce050
Fix show tables as JDBC with security enabled
This commit fixes the test SqlSecurityTestCase.testShowTablesWorksAsAdmin to
skip over any indices/aliases that exist starting with `.security`.
Use value comparison instead of the result sets
Fix an offset bug while at it (columns start at 1 not 0)
Resolveselastic/x-pack-elasticsearch#3423
Original commit: elastic/x-pack-elasticsearch@6fffda6070
This is the next step in removing the top level sql directory.
I named the directory `sql-cli` instead of `cli` because that puts it at
the maven coordinates `org.elasticsearch.plugin:sql-cli` instead of
`org.elasticsearch.plugin:cli`.
Relates to elastic/x-pack-elasticsearch#3363
Original commit: elastic/x-pack-elasticsearch@d41a57a136
This commit makes ML snapshot downloading happen less often. It does
that by first moving the download location to a directory outside the
destructive power of gradle clean, and then also uses the md5 of the zip
to compare to that found in s3. This allows us to do a cheap HEAD
request to find if the file has changed.
Original commit: elastic/x-pack-elasticsearch@cd8b00fd31
This moves SQL's server project into `plugin:sql` without modifying how the integration is performed. I know that it is not correct with regards to the x-pack modularization but I think it is a good first step.
Original commit: elastic/x-pack-elasticsearch@2f40d02e4d
The on-closing Connection is closed so there's no leak however closing
the ResultSet is good practice and tests another piece of code as well
Original commit: elastic/x-pack-elasticsearch@3a9cee70a3
This commit adds the ability to refresh tokens that have been obtained by the API using a refresh
token. Refresh tokens are one time use tokens that are valid for 24 hours. The tokens may be used
to get a new access and refresh token if the refresh token has not been invalidated or
already refreshed.
relates elastic/x-pack-elasticsearch#2595
Original commit: elastic/x-pack-elasticsearch@23435eb815
This isn't pretty but it removes our need to compile with parameter
names in the debug symbols and the use of reflection during tree
transforms. `instanceof` is still used. It does so by forcing all
subclasses of `Node` to implement two methods like this:
```
@Override
protected NodeInfo<PercentileRank, Expression> info() {
return info(this, PercentileRank::new, field(), value());
}
@Override
protected Expression replaceChildren(List<Expression> newChildren) {
if (newChildren.size() != 2) {
throw new IllegalArgumentException("Expected [2] children but got [" + newChildren.size() + "]");
}
return new PercentileRank(location(), newChildren.get(0), newChildren.get(1));
}
```
Every. Single. One.
This is tedious and painful and you have to do each one perfectly,
but it *is* checked by the compiler so it is less scary then the reflection
based approach it is replacing. Marginally. It is still pretty terrifying because
it requires so many tiny changes. While the compiler *does* check that
you've made all the right methods it doesn't check that you've implemented
them correctly.
Technically relates elastic/x-pack-elasticsearch#2871 but doesn't really close the "OO all the things" spirit
of elastic/x-pack-elasticsearch#2871.
A change like this deserves a million tests. Instead, I've created a hacky
reflection based test that attempts to verify that all subclasses of `Node`
implement these method correctly for some test verifiable definition of
"correct".
Original commit: elastic/x-pack-elasticsearch@a69ab634f4
This is related to elastic/x-pack-elasticsearch#3246. This commit adds a SSL/TLS layer to the nio
work implemented in the SSLChannelContext and SSLDriver classes.
This work is used to build up a SecurityNioTransport implementation.
This transport does yet offer feature parity with our normal security
transport. It mainly offers SSL/TLS security.
Original commit: elastic/x-pack-elasticsearch@d0e0484418
The cluster state listener used by watch now have two additional checks.
First, when no master node exists in the cluster state, watcher will
stop and the indexing listener will not try to trigger any new watch.
Second, when there is a global cluster write level block, it would not
be possible to update the watches index or write into the watcher
history, so the listener can bail at that case as well.
In addition this also changes the log level from debug to info when
watcher is stopped. It turned out that there are zero insights when or
if watcher is stopped when normal logging is activated. This makes it
super hard for support to know when watcher is stopped or started at all
due to shards being moved around.
Original commit: elastic/x-pack-elasticsearch@5e9ce24380
We need to start Gradle on the same JVM as runtime Java home to ensure
that certain tasks (e.g., third party audit) are run on all the flavors
of the JDK that we support (without this change it would only run on the
same JVM as Gradle which would usually be compiler Java home).
Relates elastic/x-pack-elasticsearch#3588
Original commit: elastic/x-pack-elasticsearch@d5f6a54fbe
* [Security] Handle cache expiry in token service
The keyCache on TokenService.KeyAndCache has a 60 minute expiry.
If the token service was idle for more than 60 minutes, the current
key would be expired and it would then fail to generate user tokens.
Original commit: elastic/x-pack-elasticsearch@fd98130a27
Jason Tedor
David Roberts
Ryan Ernst
Michael Basnight
Nik Everett
Lisa Cawley
Dimitrios Athanasiou
Lee Hinman
Lee Hinman
Costin Leau
Tim Brooks
jaymode
Costin Leau
Jay Modi
Igor Motov
Alexander Reelsen
Tim Vernum