octoleo/gh-ost
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

Merge branch 'master' into master

Browse Source
This commit is contained in:
Shlomi Noach 2018-12-25 07:20:47 +02:00 committed by GitHub
commit 419794e16d
No known key found for this signature in database
GPG Key ID: 4AEE18F83AFDEB23
4 changed files with 91 additions and 59 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

16
doc/command-line-flags.md
View File

@ -69,6 +69,10 @@ This is somewhat similar to a Nagios `n`-times test, where `n` in our case is al
Optional. Default is `safe`. See more discussion in [`cut-over`](cut-over.md)
### cut-over-lock-timeout-seconds
Default `3`. Max number of seconds to hold locks on tables while attempting to cut-over (retry attempted when lock exceeds timeout).
### discard-foreign-keys
**Danger**: this flag will _silently_ discard any foreign keys existing on your table.
@ -107,6 +111,10 @@ While the ongoing estimated number of rows is still heuristic, it's almost exact
Without this parameter, migration is a _noop_: testing table creation and validity of migration, but not touching data.
### force-table-names
Table name prefix to be used on the temporary tables.
### gcp
Add this flag when executing on a 1st generation Google Cloud Platform (GCP).
@ -125,6 +133,10 @@ We think `gh-ost` should not take chances or make assumptions about the user's t
See [`initially-drop-ghost-table`](#initially-drop-ghost-table)
### initially-drop-socket-file
Default False. Should `gh-ost` forcibly delete an existing socket file. Be careful: this might drop the socket file of a running migration!
### max-lag-millis
On a replication topology, this is perhaps the most important migration throttling factor: the maximum lag allowed for migration to work. If lag exceeds this value, migration throttles.
@ -169,6 +181,10 @@ See [`approve-renamed-columns`](#approve-renamed-columns)
Issue the migration on a replica; do not modify data on master. Useful for validating, testing and benchmarking. See [`testing-on-replica`](testing-on-replica.md)
### test-on-replica-skip-replica-stop
Default `False`. When `--test-on-replica` is enabled, do not issue commands stop replication (requires `--test-on-replica`).
### throttle-control-replicas
Provide a command delimited list of replicas; `gh-ost` will throttle when any of the given replicas lag beyond [`--max-lag-millis`](#max-lag-millis). The list can be queried and updated dynamically via [interactive commands](interactive-commands.md)

70
doc/shared-key.md
View File

@ -1,10 +1,10 @@
# Shared key
A requirement for a migration to run is that the two _before_ and _after_ tables have a shared unique key. This is to elaborate and illustrate on the matter.
gh-ost requires for every migration that both the _before_ and _after_ versions of the table share the same unique not-null key columns. This page illustrates this rule.
### Introduction
Consider a classic, simple migration. The table is any normal:
Consider a simple migration, with a normal table,
```sql
CREATE TABLE tbl (
@ -15,54 +15,72 @@ CREATE TABLE tbl (
)
```
And the migration is a simple `add column ts timestamp`.
and the migration `add column ts timestamp`. The _after_ table version would be:
In such migration there is no change in indexes, and in particular no change to any unique key, and specifically no change to the `PRIMARY KEY`. To run this migration, `gh-ost` would iterate the `tbl` table using the primary key, copy rows from `tbl` to the _ghost_ table `_tbl_gho` by order of `id`, and then apply binlog events onto `_tbl_gho`.
```sql
CREATE TABLE tbl (
id bigint unsigned not null auto_increment,
data varchar(255),
more_data int,
ts timestamp,
PRIMARY KEY(id)
)
```
Applying the binlog events assumes the existence of a shared unique key. For example, an `UPDATE` statement in the binary log translate to a `REPLACE` statement which `gh-ost` applies to the _ghost_ table. Such statement expects to add or replace an existing row based on given row data. In particular, it would _replace_ an existing row if a unique key violation is met.
(This is also the definition of the _ghost_ table, except that that table would be called `_tbl_gho`).
So `gh-ost` correlates `tbl` and `_tbl_gho` rows using a unique key. In the above example that would be the `PRIMARY KEY`.
In this migration, the _before_ and _after_ versions contain the same unique not-null key (the PRIMARY KEY). To run this migration, `gh-ost` would iterate through the `tbl` table using the primary key, copy rows from `tbl` to the _ghost_ table `_tbl_gho` in primary key order, while also applying the binlog event writes from `tble` onto `_tbl_gho`.
### Rules
The applying of the binlog events is what requires the shared unique key. For example, an `UPDATE` statement to `tbl` translates to a `REPLACE` statement which `gh-ost` applies to `_tbl_gho`. A `REPLACE` statement expects to insert or replace an existing row based on its row's values and the table's unique key constraints. In particular, if inserting that row would result in a unique key violation (e.g., a row with that primary key already exists), it would _replace_ that existing row with the new values.
There must be a shared set of not-null columns for which there is a unique constraint in both the original table and the migration (_ghost_) table.
So `gh-ost` correlates `tbl` and `_tbl_gho` rows one to one using a unique key. In the above example that would be the `PRIMARY KEY`.
### Interpreting the rules
### Interpreting the rule
The same columns must be covered by a unique key in both tables. This doesn't have to be the `PRIMARY KEY`. This doesn't have to be a key of the same name.
The _before_ and _after_ versions of the table share the same unique not-null key, but:
- the key doesn't have to be the PRIMARY KEY
- the key can have a different name between the _before_ and _after_ versions (e.g., renamed via DROP INDEX and ADD INDEX) so long as it contains the exact same column(s)
Upon migration, `gh-ost` inspects both the original and _ghost_ table and attempts to find at least one such unique key (or rather, a set of columns) that is shared between the two. Typically this would just be the `PRIMARY KEY`, but sometimes you may change the `PRIMARY KEY` itself, in which case `gh-ost` will look for other options.
At the start of the migration, `gh-ost` inspects both the original and _ghost_ table it created, and attempts to find at least one such unique key (or rather, a set of columns) that is shared between the two. Typically this would just be the `PRIMARY KEY`, but some tables don't have primary keys, or sometimes it is the primary key that is being modified by the migration. In these cases `gh-ost` will look for other options.
`gh-ost` expects unique keys where no `NULL` values are found, i.e. all columns covered by the unique key are defined as `NOT NULL`. This is implicitly true for `PRIMARY KEY`s. If no such key can be found, `gh-ost` bails out. In the event there is no such key, but you happen to _know_ your columns have no `NULL` values even though they're `NULL`-able, you may take responsibility and pass the `--allow-nullable-unique-key`. The migration will run well as long as no `NULL` values are found in the unique key's columns. Any actual `NULL`s may corrupt the migration.
`gh-ost` expects unique keys where no `NULL` values are found, i.e. all columns contained in the unique key are defined as `NOT NULL`. This is implicitly true for primary keys. If no such key can be found, `gh-ost` bails out.
### Examples: allowed and not allowed
If the table contains a unique key with nullable columns, but you know your columns contain no `NULL` values, use the `--allow-nullable-unique-key` option. The migration will run well as long as no `NULL` values are found in the unique key's columns. **Any actual `NULL`s may corrupt the migration.**
### Examples: Allowed and Not Allowed
```sql
create table some_table (
id int auto_increment,
id int not null auto_increment,
ts timestamp,
name varchar(128) not null,
owner_id int not null,
loc_id int,
loc_id int not null,
primary key(id),
unique key name_uidx(name)
)
```
Following are examples of migrations that are _good to run_:
Note the two unique, not-null indexes: the primary key and `name_uidx`.
Allowed migrations:
- `add column i int`
- `add key owner_idx(owner_id)`
- `add unique key owner_name_idx(owner_id, name)` - though you need to make sure to not write conflicting rows while this migration runs
- `add key owner_idx (owner_id)`
- `add unique key owner_name_idx (owner_id, name)` - **be careful not to write conflicting rows while this migration runs**
- `drop key name_uidx` - `primary key` is shared between the tables
- `drop primary key, add primary key(owner_id, loc_id)` - `name_uidx` is shared between the tables and is used for migration
- `change id bigint unsigned` - the `'primary key` is used. The change of type still makes the `primary key` workable.
- `drop primary key, drop key name_uidx, create primary key(name), create unique key id_uidx(id)` - swapping the two keys. `gh-ost` is still happy because `id` is still unique in both tables. So is `name`.
- `drop primary key, add primary key(owner_id, loc_id)` - `name_uidx` is shared between the tables
- `change id bigint unsigned not null auto_increment` - the `primary key` changes datatype but not value, and can be used
- `drop primary key, drop key name_uidx, add primary key(name), add unique key id_uidx(id)` - swapping the two keys. Either `id` or `name` could be used
Not allowed:
- `drop primary key, drop key name_uidx` - the _ghost_ table has no unique key
- `drop primary key, drop key name_uidx, create primary key(name, owner_id)` - no shared columns to the unique keys on both tables. Even though `name` exists in the _ghost_ table's `primary key`, it is only part of the key and in itself does not guarantee uniqueness in the _ghost_ table.
Following are examples of migrations that _cannot run_:
### Workarounds
- `drop primary key, drop key name_uidx` - no unique key to _ghost_ table, so clearly cannot run
- `drop primary key, drop key name_uidx, create primary key(name, owner_id)` - no shared columns to both tables. Even though `name` exists in the _ghost_ table's `primary key`, it is only part of the key and in itself does not guarantee uniqueness in the _ghost_ table.
Also, you cannot run a migration on a table that doesn't have some form of `unique key` in the first place, such as `some_table (id int, ts timestamp)`
If you need to change your primary key or only not-null unique index to use different columns, you will want to do it as two separate migrations:
1. `ADD UNIQUE KEY temp_pk (temp_pk_column,...)`
1. `DROP PRIMARY KEY, DROP KEY temp_pk, ADD PRIMARY KEY (temp_pk_column,...)`

8
doc/throttle.md
View File

@ -46,6 +46,14 @@ Note that you may dynamically change both `--max-lag-millis` and the `throttle-c
An example query could be: `--throttle-query="select hour(now()) between 8 and 17"` which implies throttling auto-starts `8:00am` and migration auto-resumes at `18:00pm`.
#### HTTP Throttle
The `--throttle-http` flag allows for throttling via HTTP. Every 100ms `gh-ost` issues a `HEAD` request to the provided URL. If the response status code is not `200` throttling will kick in until a `200` response status code is returned.
If no URL is provided or the URL provided doesn't contain the scheme then the HTTP check will be disabled. For example `--throttle-http="http://1.2.3.4:6789/throttle"` will enable the HTTP check/throttling, but `--throttle-http="1.2.3.4:6789/throttle"` will not.
The URL can be queried and updated dynamically via [interactive interface](interactive-commands.md).
#### Manual control
In addition to the above, you are able to take control and throttle the operation any time you like.

56
go/logic/inspect.go
View File

@ -173,8 +173,7 @@ func (this *Inspector) inspectOriginalAndGhostTables() (err error) {
// This additional step looks at which columns are unsigned. We could have merged this within
// the `getTableColumns()` function, but it's a later patch and introduces some complexity; I feel
// comfortable in doing this as a separate step.
this.applyColumnTypes(this.migrationContext.DatabaseName, this.migrationContext.OriginalTableName, this.migrationContext.OriginalTableColumns, this.migrationContext.SharedColumns)
this.applyColumnTypes(this.migrationContext.DatabaseName, this.migrationContext.OriginalTableName, &this.migrationContext.UniqueKey.Columns)
this.applyColumnTypes(this.migrationContext.DatabaseName, this.migrationContext.OriginalTableName, this.migrationContext.OriginalTableColumns, this.migrationContext.SharedColumns, &this.migrationContext.UniqueKey.Columns)
this.applyColumnTypes(this.migrationContext.DatabaseName, this.migrationContext.GetGhostTableName(), this.migrationContext.GhostTableColumns, this.migrationContext.MappedSharedColumns)
for i := range this.migrationContext.SharedColumns.Columns() {
@ -555,44 +554,35 @@ func (this *Inspector) applyColumnTypes(databaseName, tableName string, columnsL
err := sqlutils.QueryRowsMap(this.db, query, func(m sqlutils.RowMap) error {
columnName := m.GetString("COLUMN_NAME")
columnType := m.GetString("COLUMN_TYPE")
if strings.Contains(columnType, "unsigned") {
for _, columnsList := range columnsLists {
columnsList.SetUnsigned(columnName)
for _, columnsList := range columnsLists {
column := columnsList.GetColumn(columnName)
if column == nil {
continue
}
}
if strings.Contains(columnType, "mediumint") {
for _, columnsList := range columnsLists {
columnsList.GetColumn(columnName).Type = sql.MediumIntColumnType
if strings.Contains(columnType, "unsigned") {
column.IsUnsigned = true
}
}
if strings.Contains(columnType, "timestamp") {
for _, columnsList := range columnsLists {
columnsList.GetColumn(columnName).Type = sql.TimestampColumnType
if strings.Contains(columnType, "mediumint") {
column.Type = sql.MediumIntColumnType
}
}
if strings.Contains(columnType, "datetime") {
for _, columnsList := range columnsLists {
columnsList.GetColumn(columnName).Type = sql.DateTimeColumnType
if strings.Contains(columnType, "timestamp") {
column.Type = sql.TimestampColumnType
}
}
if strings.Contains(columnType, "json") {
for _, columnsList := range columnsLists {
columnsList.GetColumn(columnName).Type = sql.JSONColumnType
if strings.Contains(columnType, "datetime") {
column.Type = sql.DateTimeColumnType
}
}
if strings.Contains(columnType, "float") {
for _, columnsList := range columnsLists {
columnsList.GetColumn(columnName).Type = sql.FloatColumnType
if strings.Contains(columnType, "json") {
column.Type = sql.JSONColumnType
}
}
if strings.HasPrefix(columnType, "enum") {
for _, columnsList := range columnsLists {
columnsList.GetColumn(columnName).Type = sql.EnumColumnType
if strings.Contains(columnType, "float") {
column.Type = sql.FloatColumnType
}
}
if charset := m.GetString("CHARACTER_SET_NAME"); charset != "" {
for _, columnsList := range columnsLists {
columnsList.SetCharset(columnName, charset)
if strings.HasPrefix(columnType, "enum") {
column.Type = sql.EnumColumnType
}
if charset := m.GetString("CHARACTER_SET_NAME"); charset != "" {
column.Charset = charset
}
}
return nil