Monday, March 12, 2018

dbdeployer release candidate


The latest release of dbdeployer is possibly the last one with a leading 0. If no serious bugs are found in the next two weeks, the next release will bear a glorious 1.0.

Latest news

The decision to get out of the stream of pre-releases that were published until now comes because I have implemented all the features that I wanted to add: mainly, all the ones that I wished to add to MySQL-Sandbox but it would have been too hard:

The latest addition is the ability of running multi-source topologies. Now we can run four topologies:

  • master-slave is the default topology. It will install one master and two slaves. More slaves can be added with the option --nodes.
  • group will deploy three peer nodes in group replication. If you want to use a single primary deployment, add the option --single-primary. Available for MySQL 5.7 and later.
  • fan-in is the opposite of master-slave. Here we have one slave and several masters. This topology requires MySQL 5.7 or higher.
    all-masters is a special case of fan-in, where all nodes are masters and are also slaves of all nodes.

It is possible to tune the flow of data in multi-source topologies. The default for fan-in is three nodes, where 1 and 2 are masters, and 2 are slaves. You can change the predefined settings by providing the list of components:

$ dbdeployer deploy replication \
    --topology=fan-in \
    --nodes=5 \
    --master-list="1 2 3" \
    --slave-list="4 5" \
    8.0.4 \
    --concurrent

In the above example, we get 5 nodes instead of 3. The first three are master (--master-list="1 2 3") and the last two are slaves (--slave-list="4 5") which will receive data from all the masters. There is a test automatically generated to test replication flow. In our case it shows the following:

$ ~/sandboxes/fan_in_msb_8_0_4/test_replication
# master 1
# master 2
# master 3
# slave 4
ok - '3' == '3' - Slaves received tables from all masters
# slave 5
ok - '3' == '3' - Slaves received tables from all masters
# pass: 2
# fail: 0

The first three lines show that each master has done something. In our case, each master has created a different table. Slaves in nodes 5 and 6 then count how many tables they found, and if they got the tables from all masters, the test succeeds.
Note that for all-masters topology there is no need to specify master-list or slave-list. In fact, those lists will be auto-generated, and they will both include all deployed nodes.

What now?

Once I make sure that the current features are reasonably safe (I will only write more tests for the next 10~15 days) I will publish the first (non-pre) release of dbdeployer. From that moment, I'd like to follow the recommendations of the Semantic Versioning:

  • The initial version will be 1.0.0 (major, minor, revision);
  • The spects for 1.0 will be the API that needs to be maintained.
  • Bug fixes will increment the revision counter.
  • New features that don't break compatibility with the API will increment the minor counter;
  • New features or changes that break compatibility will trigger a major counter increment.

Using this method will give users a better idea of what to expect. If we get a revision number increase, it is only bug fixes. An increase in the minor counter means that there are new features, but all previous features work as before. An increase in the major counter means that something will break, either because of changed interface or because of changed behavior.
In practice, the tests released with 1.0.0 should run with any 1.x subsequent version. When those tests need changes to run correctly, we will need to bump up the major version.

Let's see if this method is sustainable. So far, I haven't had need to do behavioural changes, which are usually provoked by new versions of MySQL that introduce incompatible behavior (definitely MySQL does not follow the Semantic Versioning principles.) When the next version becomes available, I will see if this RC of dbdeployer can stand its ground.


Sunday, March 11, 2018

Concurrent sandbox deployment


Version 0.3.0 of dbdeployer has gained the ability of deploying multiple sandboxes concurrently. Whenever we deploy a group of sandboxes (replication, multiple) we can use the --concurrent flag, telling dbdeployer that it should run operations concurrently.

What happens when a single sandbox gets deployed? There are six sets of operations:

  1. Create the sandbox directory and write down its scripts;
  2. Run the initialisation script;
  3. Start the database server;
  4. Run the pre-grants SQL commands (if any;)
  5. Load the grants;
  6. Run the post-grants SQL commands (if any;)

When several sandboxes are deployed concurrently, dbdeployer runs only the first step, and then creates a list of commands with an associated priority index. These commands are assembled for every sandbox, and then executed concurrently for every step.
The sequence of events for a deployment of three sandboxes in replication would be like this:

  1. Create the sandbox skeleton for every sandbox;
  2. Initialise all database servers;
  3. start all the servers;
  4. run the pre-grants, grants, post-grants scripts.
  5. Runs the group initialisation script (start master and slaves, or setup group replication).

Depending on the computer architecture, the server version, and the number of nodes, the speed of deployment can increase from 2 to 5 times.

Let's see an example:

$ time dbdeployer deploy replication 5.7.21
[...]
real    0m13.789s
user    0m1.143s
sys 0m1.873s

$ time dbdeployer deploy replication 5.7.21 --concurrent
[...]
real    0m7.780s
user    0m1.329s
sys 0m1.811s

There is a significant speed increase. The gain rises sharply if we use an higher number of nodes.

$ time dbdeployer deploy replication 5.7.21 --nodes=5
[...]
real    0m23.425s
user    0m1.923s
sys 0m3.106s

$ time dbdeployer deploy replication 5.7.21 \
    --nodes=5 --concurrent
[...]
real    0m7.686s
user    0m2.248s
sys 0m2.777s

As we can see, the time for deploying 5 nodes is roughly the same used for 3 nodes. While the sequential operations take time proportionally with the number of nodes, the concurrent task stays almost constant.

Things a re a bit different for group replication, as the group initialisation (which happens after all the servers are up and running) takes more time than the simple master/slave deployment, and can't be easily reduced using the current code.

A similar optimisation happens when we delete multiple sandboxes. Here the operation is at sandbox level (1 replication cluster = 1 sandbox) not at server level, and for that reason the gain is less sharp. Still, operations are noticeably faster.

There is room for improvement, but I have seen that the total testing time for dbdeployer test suite has dropped from 26 to 15 minutes. I think it was a week end well spent.

Monday, March 05, 2018

Customizing dbdeployer


As of version 0.2.1, dbdeployer allows users to customize composite sandboxes more than ever. This is done by manipulating the default settings, which are used to deploy the sandbox templates.

In order to appreciate the customization capabilities, let's start with a vanilla deployment, and then we have a look at the possible changes.

$ dbdeployer deploy replication 8.0.4
Installing and starting master
Database installed in $HOME/sandboxes/rsandbox_8_0_4/master
. sandbox server started
Installing and starting slave 1
Database installed in $HOME/sandboxes/rsandbox_8_0_4/node1
. sandbox server started
Installing and starting slave 2
Database installed in $HOME/sandboxes/rsandbox_8_0_4/node2
. sandbox server started
$HOME/sandboxes/rsandbox_8_0_4/initialize_slaves
initializing slave 1
initializing slave 2
Replication directory installed in $HOME/sandboxes/rsandbox_8_0_4
run 'dbdeployer usage multiple' for basic instructions'

A regular replication sandbox has one master and two slaves. Each slave is inside a directory called nodeX.

The resulting sandbox has a directory called master, two nodeX directories, a shortcut for the master called m, and two shortcuts for the slaves called s1 and s2. There are also two management scripts called initialize_slaves and check_slaves.

    $ ls -l ~/sandboxes/rsandbox_8_0_4/
    total 152
    -rwxr--r--   1 user  staff  1500 Mar  5 06:21 check_slaves
    -rwxr--r--   1 user  staff  1160 Mar  5 06:21 clear_all
    -rwxr--r--   1 user  staff  1617 Mar  5 06:21 initialize_slaves
    -rwxr--r--   1 user  staff   806 Mar  5 06:21 m
    drwxr-xr-x  22 user  staff   748 Mar  5 06:21 master
    -rwxr--r--   1 user  staff   806 Mar  5 06:21 n1
    -rwxr--r--   1 user  staff   804 Mar  5 06:21 n2
    -rwxr--r--   1 user  staff   804 Mar  5 06:21 n3
    drwxr-xr-x  23 user  staff   782 Mar  5 06:21 node1
    drwxr-xr-x  23 user  staff   782 Mar  5 06:21 node2
    -rwxr--r--   1 user  staff   855 Mar  5 06:21 restart_all
    -rwxr--r--   1 user  staff   804 Mar  5 06:21 s1
    -rwxr--r--   1 user  staff   804 Mar  5 06:21 s2
    -rw-r--r--   1 user  staff   173 Mar  5 06:21 sbdescription.json
    -rwxr--r--   1 user  staff  1127 Mar  5 06:21 send_kill_all
    -rwxr--r--   1 user  staff  1296 Mar  5 06:21 start_all
    -rwxr--r--   1 user  staff  1680 Mar  5 06:21 status_all
    -rwxr--r--   1 user  staff  1087 Mar  5 06:21 stop_all
    -rwxr--r--   1 user  staff  4598 Mar  5 06:21 test_replication
    -rwxr--r--   1 user  staff  1315 Mar  5 06:21 test_sb_all
    -rwxr--r--   1 user  staff  1100 Mar  5 06:21 use_all

Now, let's see how we can change this. We'll start by listing the current defaults

$ dbdeployer defaults show
# Internal values:
{
    "version": "0.2.1",
    "sandbox-home": "$HOME/sandboxes",
    "sandbox-binary": "$HOME/opt/mysql",
    "master-slave-base-port": 11000,
    "group-replication-base-port": 12000,
    "group-replication-sp-base-port": 13000,
    "fan-in-replication-base-port": 14000,
    "all-masters-replication-base-port": 15000,
    "multiple-base-port": 16000,
    "group-port-delta": 125,
    "master-name": "master",
    "master-abbr": "m",
    "node-prefix": "node",
    "slave-prefix": "slave",
    "slave-abbr": "s",
    "sandbox-prefix": "msb_",
    "master-slave-prefix": "rsandbox_",
    "group-prefix": "group_msb_",
    "group-sp-prefix": "group_sp_msb_",
    "multiple-prefix": "multi_msb_",
    "fan-in-prefix": "fan_in_msb_",
    "all-masters-prefix": "all_masters_msb_"
 }

The values that we want to change are master-name, master-abbr, node-prefix, slave-prefix, and slave-abbr. We can export the defaults to a file, and import them after editing the values we want to change.

$ dbdeployer defaults export defaults.json
# Defaults exported to file defaults.json
$ vim defaults.json
$ dbdeployer defaults import defaults.json
Defaults imported from defaults.json into $HOME/.dbdeployer/config.json

Now dbdeployer is using the new defaults.


$ dbdeployer defaults show
# Configuration file: $HOME/.dbdeployer/config.json
{
    "version": "0.2.1",
    "sandbox-home": "/Users/gmax/sandboxes",
    "sandbox-binary": "/Users/gmax/opt/mysql",
    "master-slave-base-port": 11000,
    "group-replication-base-port": 12000,
    "group-replication-sp-base-port": 13000,
    "fan-in-replication-base-port": 14000,
    "all-masters-replication-base-port": 15000,
    "multiple-base-port": 16000,
    "group-port-delta": 125,
    "master-name": "primary",
    "master-abbr": "p",
    "node-prefix": "branch",
    "slave-prefix": "replica",
    "slave-abbr": "r",
    "sandbox-prefix": "msb_",
    "master-slave-prefix": "rsandbox_",
    "group-prefix": "group_msb_",
    "group-sp-prefix": "group_sp_msb_",
    "multiple-prefix": "multi_msb_",
    "fan-in-prefix": "fan_in_msb_",
    "all-masters-prefix": "all_masters_msb_"
 }
We have now *primary* for *master*, *replica* for *slave*, *branch* for *node*, and the abbreviations for master and slave changed to *p* and *r* respectively.
Let's see how these defaults can play together when we run the same command as we did before for replication. We first remove the previous deployment.

$ dbdeployer delete rsandbox_8_0_4
List of deployed sandboxes:
$HOME/sandboxes/rsandbox_8_0_4
Running $HOME/sandboxes/rsandbox_8_0_4/stop_all
# executing "stop" on $HOME/sandboxes/rsandbox_8_0_4
executing "stop" on slave 1
executing "stop" on slave 2
executing "stop" on master
Running rm -rf $HOME/sandboxes/rsandbox_8_0_4
Sandbox $HOME/sandboxes/rsandbox_8_0_4 deleted

The deployment command is the same as before, but the output changes:

$ dbdeployer deploy replication 8.0.4
Installing and starting primary
Database installed in $HOME/sandboxes/rsandbox_8_0_4/primary
. sandbox server started
Installing and starting replica 1
Database installed in $HOME/sandboxes/rsandbox_8_0_4/branch1
. sandbox server started
Installing and starting replica 2
Database installed in $HOME/sandboxes/rsandbox_8_0_4/branch2
.. sandbox server started
$HOME/sandboxes/rsandbox_8_0_4/initialize_replicas
initializing replica 1
initializing replica 2
Replication directory installed in $HOME/sandboxes/rsandbox_8_0_4
run 'dbdeployer usage multiple' for basic instructions'

This looks already as if our defaults have been adopted. Let's see the sandbox itself:

$ ls -l ~/sandboxes/rsandbox_8_0_4/
total 152
drwxr-xr-x  23 user  staff   782 Mar  5 06:45 branch1
drwxr-xr-x  23 user  staff   782 Mar  5 06:45 branch2
-rwxr--r--   1 user  staff  1515 Mar  5 06:45 check_replicas
-rwxr--r--   1 user  staff  1170 Mar  5 06:45 clear_all
-rwxr--r--   1 user  staff  1629 Mar  5 06:45 initialize_replicas
-rwxr--r--   1 user  staff   807 Mar  5 06:45 n1
-rwxr--r--   1 user  staff   806 Mar  5 06:45 n2
-rwxr--r--   1 user  staff   806 Mar  5 06:45 n3
-rwxr--r--   1 user  staff   807 Mar  5 06:45 p
drwxr-xr-x  22 user  staff   748 Mar  5 06:45 primary
-rwxr--r--   1 user  staff   806 Mar  5 06:45 r1
-rwxr--r--   1 user  staff   806 Mar  5 06:45 r2
-rwxr--r--   1 user  staff   855 Mar  5 06:45 restart_all
-rw-r--r--   1 user  staff   173 Mar  5 06:45 sbdescription.json
-rwxr--r--   1 user  staff  1137 Mar  5 06:45 send_kill_all
-rwxr--r--   1 user  staff  1308 Mar  5 06:45 start_all
-rwxr--r--   1 user  staff  1700 Mar  5 06:45 status_all
-rwxr--r--   1 user  staff  1097 Mar  5 06:45 stop_all
-rwxr--r--   1 user  staff  4613 Mar  5 06:45 test_replication
-rwxr--r--   1 user  staff  1325 Mar  5 06:45 test_sb_all
-rwxr--r--   1 user  staff  1106 Mar  5 06:45 use_all

We see that the new defaults were used and the script names have changed. But the differences are deeper than this. Also the internal values in the scripts were changed accordingly.

$ ~/sandboxes/rsandbox_8_0_4/test_replication
# primary log: mysql-bin.000001 - Position: 14073 - Rows: 20
# Testing replica #1
ok - replica #1 acknowledged reception of transactions from primary
ok - replica #1 IO thread is running
ok - replica #1 SQL thread is running
ok - Table t1 found on replica #1
ok - Table t1 has 20 rows on #1
# Testing replica #2
ok - replica #2 acknowledged reception of transactions from primary
ok - replica #2 IO thread is running
ok - replica #2 SQL thread is running
ok - Table t1 found on replica #2
ok - Table t1 has 20 rows on #2
# Tests :    10
# failed:     0 (  0.0%)
# PASSED:    10 (100.0%)
# exit code: 0

The test script calls the components with the names that we defined in the new defaults. Let's have a look at what the shortcuts for the master and slaves (now primary and replicas) do:

$ ~/sandboxes/rsandbox_8_0_4/p
Welcome to the MySQL monitor.  Commands end with ; or \g.
Your MySQL connection id is 35
Server version: 8.0.4-rc-log MySQL Community Server (GPL)

Copyright (c) 2000, 2018, Oracle and/or its affiliates. All rights reserved.

Oracle is a registered trademark of Oracle Corporation and/or its
affiliates. Other names may be trademarks of their respective
owners.

Type 'help;' or '\h' for help. Type '\c' to clear the current input statement.

primary [localhost] {msandbox} ((none)) >

$ ~/sandboxes/rsandbox_8_0_4/r1
Welcome to the MySQL monitor.  Commands end with ; or \g.
Your MySQL connection id is 15
Server version: 8.0.4-rc-log MySQL Community Server (GPL)

Copyright (c) 2000, 2018, Oracle and/or its affiliates. All rights reserved.

Oracle is a registered trademark of Oracle Corporation and/or its
affiliates. Other names may be trademarks of their respective
owners.

Type 'help;' or '\h' for help. Type '\c' to clear the current input statement.

replica1 [localhost] {msandbox} ((none)) >

Also the internal prompt has been adapted to the new naming.

Should we want to revert to the old behavior, we can just reset the defaults:

$ dbdeployer defaults reset
#File $HOME/.dbdeployer/config.json removed

The current replication sandbox is left untouched, but the next one will use the default values.

If we don't want to change the defaults permanently, there is an alternative. The --defaults flag allows us to change defaults on-the-fly just for the command we're running. For example, we could have achieved the same result, without editing the configuration file, using this command:

    dbdeployer deploy replication 8.0.4 \
        --defaults=master-name:primary \
        --defaults=master-abbr:p \
        --defaults=slave-prefix:replica \
        --defaults=slave-abbr:r \
        --defaults=node-prefix:branch

The syntax for --defaults requires the name of the variable and the new value, separated by a colon. The flag can be used as many times as needed.

MySQL security for real users


Security features overview

One of Oracle's tenets is the focus on security. For this reason, when it took over the stewardship of MySQL, it started addressing the most common issues. It was not quick acting, but we have seen real progress:

  1. MySQL 5.7 has removed the anonymous accounts, which was the greatest threat to security. Because of those accounts, and the default privileges granted to them, users without any privileges could access the "test" database and do serious damage. Additionally, because of the way the privilege engine evaluates accounts, anonymous users could hijack legitimate users, by preventing them to work properly.
  2. The "root" account now comes with a password defined during initialization. This is good news for security, but bad news for how the change was implemented.
  3. There is a new way of setting an options file for connection credentials: the mysql_config_editor paired with option --login-path allows users to store encrypted credentials for secure use. Also here, while we should rejoice for the added security, we can't help feeling that the implementation is yet again far from meeting users needs.
  4. There is an useful warning (introduced in MySQL 5.6) when using a password on the command line, telling users that it is a risk. Also in this case, we have a usability issue: while users care about their production deployments and use option files to avoid using passwords on the command line, there are, nonetheless, a lot of testing scripts, used in safe environment or with non-valuable data, where a password in the command line was not an issue, and the new warning simply screws up the result of those carefully crafted tests. This change, which can't be overcome without modifying the MySQL clients code, needs users to change their existing tests to adapt to the new behavior.
  5. MySQL 8 introduces roles, which simplify the accounts management. There are some minor usability issues, although in general the feature meets expectations.

This is the scenario of the main enhancements in MySQL since 5.6. Each one of them has some usability problems, some minor, some really bad.
We will first have a look at the problems mentioned above, and then examine the root cause for why they have arisen.


Usability issues

I start by noticing that some developers in the MySQL team have been working there for many years, starting with the time when MySQL was a different database and was used really differently.
In those times, managing the database meant that a human (the DBA) would run operations manually, take a look at the result, and adjust when needed. And then, when things went wrong, the same human explored the database system to find out what happened, took action, and went back to sleep.

Human-centered management leads to human problems: lazy DBA left their databases without password, using the root account, and exposing the server to uninspired attacks; they used passwords on the command line, without caring for options files (or without knowing about them.) Careless DBAs did not deal with anonymous users, leaving a dangerous backdoor in their server.

Some of the new functionalities introduced in the latest MySQL versions are aimed at this type of users: when you install MySQL, you get a message saying: your root password is ************, and the lazy DBAs have no option but to take note and use it. When they use the password on the command line, the annoying warning forces them to start using an options file or the mysql_config_editor.

This is all good, but the main problem here is that the DBAs of 10 years ago are on the verge of extinction. They are replaced by a new breed of DBAs who are not lazy, because they can't afford to be, and need to use dozens, hundreds, thousands of databases at once, using configuration management tools that don't require manual intervention, and actually abhor it. In the land of automation, some of the MySQL security enhancements are not seen as a solution, but as new problems.

Let's see an interesting example: docker containers.

Using Docker, MySQL images are deployed using a password on the command line. This is done for compatibility with the first implementation of the image maintained by the Docker team, where you deploy with this syntax:

docker run -e MYSQL_ROOT_PASSWORD=secret -d mysql

The MYSQL_ROOT_PASSWORD is a directive that becomes an environment variable inside the container, and the server uses it during initialization. As you can imagine, this is not recommended for a secure environment. Then, what's the MySQL team recommendation? They suggest the same strategy used for manual installation: set a directive MYSQL_RANDOM_ROOT_PASSWORD that results in a random password being generated, then collected by the DBA and used. Alternatively, the directive MYSQL_ONETIME_PASSWORD will force the root user to change the password on first connection.

The above suggestions were designed with the ancient DBA still in mind, while container deployment is even more automated than VMs, and it is based on the principle of immutable objects, i.e. containers that spring up from the cloud ready to run, with no configuration needed, and especially no configuration that requires someone (or some tool) to extract a new password from a log. I proposed a different solution, that would never show passwords on the command line and while it was implemented, but it still feels like a hack to circumvent an inadequate design.

As a result, the implementation inside the MySQL recommended Docker image uses "--initialize-insecure" to start the server. This is an implicit recognition of the bad design of the initialization feature. What was designed to overcome DBA's laziness becomes an obstacle towards automation.

We have a similar problem with mysql_config_editor: the tool will create a safe configuration file with credentials for multiple instances, but the password must be inserted manually. Consequently, this potentially useful feature doesn't get adopted, because it would be too difficult or impossible to automate properly.

We have seen that, of the security features that were introduced lately, only a few can be used safely in an automated environment, and all of them have at least one small usability quirk. I have talked about a confusing issue related to the removal of anonymous users where in their eagerness of removing the vulnerability the MySQL team removed also the "test" database, which was a consequence, not the cause of the problem. And I have recently talked about roles usability where there are still open problems, like the ability of telling roles from users which are apparently not considered a bug by the MySQL team.

All the above considerations led me to ask: how did we get to this point? There is an active community, and feedback is offered often with plenty of detail. How come we have such an abundance of usability issues? Don't the developers spend time with users at conferences to learn what they expect? Don't they read articles and blog posts about how a new feature meets expectations? Don't they talk to customers who have adopted new features? They certainly do. Then, why the usability problems persist?

What follows is my observation and speculation on this matter.


Disconnection between MySQL developers and users community

My experience working with system providers has put me in contact with many users. I have seen that in most cases users are very much protective of their current deployment, because it took them long time to get it right, and they don't upgrade unless they don't have another choice. I've seen users test the newer versions, realize that they would break some of their procedures, and defer the upgrade to better times that never come. I remember last year a user with a not so large set of servers was considering an upgrade to MySQL 5.6, while 5.7 had been GA for two years. The reason was a set of incompatibilities that made the upgrade too difficult.

For companies that deal with thousands of servers, the problem is similar, but exacerbated by the quantity of servers to upgrade and the need to do it without stopping operations. This latest requirement has made some users decide not to use GTID, because it required offline time for a master, and they hadn't had time enough to test the upgrade to MySQL 5.7 that would solve that problem.

For one reason or the other, many companies upgrade only two or three years after a given version became GA. And this is the main problem: until they use it in production, or at least test the version for a projected upgrade, users can't give valuable feedback, the one that is related to usage in production, and when they do, the version for which they provide feedback has been GA for long time, and can't be changed, while the next one is already close to GA, and as such will be untouchable.

The MySQL team gets feedback on a release from a handful of curious users who don't delay testing until the new version is GA, but don't provide the kind of important feedback that get the development team attention, such as deployment in production by large customers. In many cases, large customers are the ones that upgrade several years after GA, and by then their input is difficult to act upon.

We have then a curious situation, where the warnings given by the early software explorers are confirmed years later by the users to which the MySQL team listens more closely, but by then the next version of the server is already locked in a release schedule that nobody wants to alter to fix usability issues.

How can we solve this problem? Simple: listen to early software explorers and try to fix problems before GA.

Thursday, March 01, 2018

Using MySQL 8.0: what to expect

Mysql8

MySQL 8.0 will be GA soon (just my assumption: Oracle doesn't tell me anything about its release plans) and it's time to think about having a look at it.
If this is your first try of MySQL 8, get prepared for several impacting differences from previous versions.

In this article I won't tell you what you can do with MySQL 8: there is plenty of material about this, including in this very blog. I will instead concentrate on differences from previous versions that users need to know if they want to avoid surprises.

Data Directory

Let's start with an observation of the data directory.
After a standard installation, without any additional options, I see the following:

Files that I expected to see

auto.cnf
ib_buffer_pool
ib_logfile0
ib_logfile1
ibdata1
ibtmp1
(dir) mysql
(dir) performance_schema
(dir) sys

These files are also present in 5.7.

Files that are new in 8.0

binlog.000001
binlog.index

log-bin is ON by default. You need to remember this if you are using a MySQL server for a benchmark test that used to run without binary logs.

ca-key.pem
ca.pem
client-cert.pem
client-key.pem
private_key.pem
public_key.pem
server-cert.pem
server-key.pem

Now the MySQL generates all the certificates needed to run connections securely. This will greatly simplify your task when setting up a new instance.

mysql.ibd

This was completely unexpected! The mysql database has now its own tablespace. This is probably due to the new Data Dictionary, which is implemented in InnoDB. You will notice that all the InnoDB tables in MySQL use this tablespace, not only dictionary tables. This will help keeping administrative data separate from operational data in the rest of the server.

undo_001
undo_002

The undo logs have now their own tablespace by default.

Global variables

There are a lot of changes in global variables. Here's the list of what will impact your work when you use MySQL 8.0 for the first time:

character_set_client        utf8mb4
character_set_connection    utf8mb4
character_set_database      utf8mb4
character_set_results       utf8mb4
character_set_server        utf8mb4

All character sets are now utf8mb4. In MySQL 5.7, the default values are a mix of utf8 and latin1.

default_authentication_plugin   caching_sha2_password

This is huge. Using this plugin, passwords are stored in a different way, which guarantees more security, but will probably break several workflows among the users. The bad thing about this change implementation is that this password format contains characters that don't display well on screen, and you can see garbled output when inspecting the "user" table.

local_infile    OFF

Loading local files is now prevented by default. If you have a workflow that requires such operations, you need to enable it.

log_bin ON
log_slave_updates   ON

We've seen from an inspection of the local directory that binary logging is enabled by default. But also very important is that log_slave_update is enabled. This is important to have slaves ready to replace a master, but will severely affect performance in those scenarios where some slaves were supposed to run without that feature.

master_info_repository  TABLE
relay_log_info_repository   TABLE

Also impacting performance is the setting for replication repositories, which are now on TABLE by default. This is something that should have happened already in MySQL 5.6 and was long overdue.

Surprisingly, something that DOES NOT get enabled by default is Global Transaction Identifiers (GTID). This is also a legacy from decisions taken in MySQL 5.6. Due to the GTID implementation, enabling them by default is not possible when upgrading from a previous version. With new data in a fresh installation, it is safe to enable GTID from the start.


Users


There are two new users when the server is created:

mysql.infoschema
mysql.session 

Theoretically, mysql.session also exists in 5.7, but it was introduced long after GA, so it still qualifies as a novelty.

Then, when the server starts, you get a grand total of 4 users (root and mysql.sys are inherited from MySQL 5.7.)


Mixed oddities


When MySQL initializes, i.e. when the server starts for the first time and creates the database, you will notice some slowness, compared to previous versions. This is in part due to the data dictionary, which needs to create and fill 30 tables, but it is not a big deal in terms of performance. In some systems, though, the slowness is so acute that you start worrying about the server being stuck.

I noticed this problem in my Intel NUC running with SSD storage. In this box, the initialization time took a serious hit:

Version time
5.0.96 1.231s
5.1.72 1.346s
5.5.52 2.441s
5.6.39 5.540s
5.7.21 6.080s
8.0.3 7.826s
8.0.4 38.547s

There is no mistype. The initialization for 8.0.4 lasts 6 times more than 5.7.
This doesn't happen everywhere. On a Mac laptop running on SSD the same operation takes almost 9 seconds, while 5.7 deploys in less than 5. It is still a substantial difference, one that has totally disrupted my regular operations in the NUC. I investigated the matter, and I found the reason. In 8.0, we have a new (hidden) table in the data dictionary, called st_spatial_reference_systems. Up to MySQL 8.0.3, this table was filled using a single transaction containing roughly 5,000 REPLACE INTO statements. It is a lot of data, but it happens quickly. For comparison, in MySQL 8.0.3 the initialization is only 2 seconds slower than 5.7.
The reason for the slowness in 8.0.4 is that there was a new command added to the syntax: CREATE SPATIAL REFERENCE SYSTEM, which is now used 5,000 times to fill the table that was previously filled with a single transaction. I don't know why someone in the MySQL team thought that changing this operation that is hidden from users was a good idea. The data is contained in the server itself and it goes into a data dictionary table, also not visible to users. I am sure I can find at least two methods to load the data faster. I was told that this glitch will be fixed in the next release. I'm waiting.

Speaking of initialization, the mysql_install_db script has been removed for good in 8.0. If you are still using it instead of the recommended mysqld --initialize, you should adapt asap.

This list is far from being exhaustive. I recommend reading What's new in MySQL 8 before upgrading.
If you are impatient, dbdeployer can help you test MySQL 8 quickly and safely.