minio/cmd/prepare-storage.go
Harshavardhana b9fc4150f6 Fix preInit logic when mixed disk situations exist. (#4904)
When servers are started simultaneously across multiple
nodes or simulating a local setup, it can happen such
that one of the servers in setup reaches a following
situation where it observes

 - Some servers are formatted
 - Some servers are unformatted
 - Some servers are offline

Current state machine doesn't handle this correctly, to fix
this situation where we have unformatted, formatted and
disks offline we do not decisively know the course of
action. So we wait for the offline disks to change their state.

Once the offline disks change their state to either one of these
states we can decisively move forward.

  - nil (formatted disk)
  - errUnformattedDisk
  - Or any other error such as errCorruptedDisk.

Fixes #4903
2017-09-12 12:17:44 -07:00

367 lines
12 KiB
Go

/*
* Minio Cloud Storage, (C) 2016 Minio, Inc.
*
* Licensed under the Apache License, Version 2.0 (the "License");
* you may not use this file except in compliance with the License.
* You may obtain a copy of the License at
*
* http://www.apache.org/licenses/LICENSE-2.0
*
* Unless required by applicable law or agreed to in writing, software
* distributed under the License is distributed on an "AS IS" BASIS,
* WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
* See the License for the specific language governing permissions and
* limitations under the License.
*/
package cmd
import (
"errors"
"fmt"
"time"
"github.com/minio/mc/pkg/console"
)
/*
Following table lists different possible states the backend could be in.
* In a single-node, multi-disk setup, "Online" would refer to disks' status.
* In a multi-node setup, it could refer to disks' or network connectivity
between the nodes, or both.
+----------+--------------------------+-----------------------+
| Online | Format status | Course of action |
| | | |
-----------+--------------------------+-----------------------+
| All | All Formatted | |
+----------+--------------------------+ initObjectLayer |
| Quorum | Quorum Formatted | |
+----------+--------------------------+-----------------------+
| All | Quorum | Print message saying |
| | Formatted, | "Heal via control" |
| | some unformatted | and initObjectLayer |
+----------+--------------------------+-----------------------+
| All | None Formatted | FormatDisks |
| | | and initObjectLayer |
| | | |
+----------+--------------------------+-----------------------+
| No | | Wait till enough |
| Quorum | _ | nodes are online and |
| | | one of the above |
| | | sections apply |
+----------+--------------------------+-----------------------+
| | | |
| Quorum | Quorum UnFormatted | Abort |
+----------+--------------------------+-----------------------+
A disk can be in one of the following states.
- Unformatted
- Formatted
- Corrupted
- Offline
*/
// InitActions - a type synonym for enumerating initialization activities.
type InitActions int
const (
// FormatDisks - see above table for disk states where it is applicable.
FormatDisks InitActions = iota
// SuggestToHeal - Prints heal message and initialize object layer.
SuggestToHeal
// WaitForQuorum - Wait for quorum number of disks to be online.
WaitForQuorum
// WaitForAll - Wait for all disks to be online.
WaitForAll
// WaitForFormatting - Wait for formatting to be triggered
// from the '1st' server in the cluster.
WaitForFormatting
// WaitForConfig - Wait for all servers to have the same config
// including (credentials, version and time).
WaitForConfig
// InitObjectLayer - Initialize object layer.
InitObjectLayer
// Abort initialization of object layer since there aren't enough good
// copies of format.json to recover.
Abort
)
// Quick error to actions converts looking for specific errors
// which need to be returned quickly and server should wait instead.
func quickErrToActions(errMap map[error]int) InitActions {
var action InitActions
switch {
case errMap[errInvalidAccessKeyID] > 0:
fallthrough
case errMap[errAuthentication] > 0:
fallthrough
case errMap[errServerVersionMismatch] > 0:
fallthrough
case errMap[errServerTimeMismatch] > 0:
action = WaitForConfig
}
return action
}
// Preparatory initialization stage for XL validates known errors.
// Converts them into specific actions. These actions have special purpose
// which caller decides on what needs to be done.
// Logic used in this function is as shown below.
//
// ---- Possible states and handled conditions -----
//
// - Formatted setup
// - InitObjectLayer when `disksFormatted >= readQuorum`
// - Wait for quorum when `disksFormatted < readQuorum && disksFormatted + disksOffline >= readQuorum`
// (we don't know yet if there are unformatted disks)
// - Wait for heal when `disksFormatted >= readQuorum && disksUnformatted > 0`
// (here we know there is at least one unformatted disk which requires healing)
//
// - Unformatted setup
// - Format/Wait for format when `disksUnformatted == diskCount`
//
// - Wait for all when `disksUnformatted + disksFormatted + diskOffline == diskCount`
//
// Under all other conditions should lead to server initialization aborted.
func prepForInitXL(firstDisk bool, sErrs []error, diskCount int) InitActions {
// Count errors by error value.
errMap := make(map[error]int)
for _, err := range sErrs {
errMap[errorCause(err)]++
}
// Validates and converts specific config errors into WaitForConfig.
if quickErrToActions(errMap) == WaitForConfig {
return WaitForConfig
}
readQuorum := diskCount / 2
disksOffline := errMap[errDiskNotFound]
disksFormatted := errMap[nil]
disksUnformatted := errMap[errUnformattedDisk]
// No Quorum lots of offline disks, wait for quorum.
if disksOffline > readQuorum {
return WaitForQuorum
}
// All disks are unformatted, proceed to formatting disks.
if disksUnformatted == diskCount {
// Only the first server formats an uninitialized setup, others wait for notification.
if firstDisk { // First node always initializes.
return FormatDisks
}
return WaitForFormatting
}
// Already formatted and in quorum, proceed to initialization of object layer.
if disksFormatted >= readQuorum {
if disksFormatted+disksOffline == diskCount {
return InitObjectLayer
}
// Some of the formatted disks are possibly corrupted or unformatted,
// let user know to heal them.
return SuggestToHeal
}
// Some unformatted, some disks formatted and some disks are offline but we don't
// quorum to decide. This is an undecisive state - wait for all of offline disks
// to be online to figure out the course of action.
if disksUnformatted+disksFormatted+disksOffline == diskCount {
return WaitForAll
}
// Exhausted all our checks, un-handled situations such as some disks corrupted we Abort.
return Abort
}
// Prints retry message upon a specific retry count.
func printRetryMsg(sErrs []error, storageDisks []StorageAPI) {
for i, sErr := range sErrs {
switch sErr {
case errDiskNotFound, errFaultyDisk, errFaultyRemoteDisk:
errorIf(sErr, "Disk %s is still unreachable", storageDisks[i])
}
}
}
// Maximum retry attempts.
const maxRetryAttempts = 5
// Implements a jitter backoff loop for formatting all disks during
// initialization of the server.
func retryFormattingXLDisks(firstDisk bool, endpoints EndpointList, storageDisks []StorageAPI) error {
if len(endpoints) == 0 {
return errInvalidArgument
}
if storageDisks == nil {
return errInvalidArgument
}
// Done channel is used to close any lingering retry routine, as soon
// as this function returns.
doneCh := make(chan struct{})
// Indicate to our retry routine to exit cleanly, upon this function return.
defer close(doneCh)
// prepare getElapsedTime() to calculate elapsed time since we started trying formatting disks.
// All times are rounded to avoid showing milli, micro and nano seconds
formatStartTime := time.Now().Round(time.Second)
getElapsedTime := func() string {
return time.Now().Round(time.Second).Sub(formatStartTime).String()
}
// Wait on the jitter retry loop.
retryTimerCh := newRetryTimerSimple(doneCh)
for {
select {
case retryCount := <-retryTimerCh:
// Attempt to load all `format.json` from all disks.
formatConfigs, sErrs := loadAllFormats(storageDisks)
if retryCount > maxRetryAttempts {
// After max retry attempts we start printing
// actual errors for disks not being available.
printRetryMsg(sErrs, storageDisks)
}
// Pre-emptively check if one of the formatted disks
// is invalid. This function returns success for the
// most part unless one of the formats is not consistent
// with expected XL format. For example if a user is
// trying to pool FS backend into an XL set.
if index, err := checkFormatXLValues(formatConfigs); err != nil {
// We will perhaps print and retry for the first 5 attempts
// because in XL initialization first server is the one which
// initializes the erasure set. This check ensures that the
// rest of the other servers do get a chance to see that the
// first server has a wrong format and exit gracefully.
// refer - https://github.com/minio/minio/issues/4140
if retryCount > maxRetryAttempts {
errorIf(err, "%s : Detected disk in unexpected format",
storageDisks[index])
continue
}
return err
}
// Check if this is a XL or distributed XL, anything > 1 is considered XL backend.
switch prepForInitXL(firstDisk, sErrs, len(storageDisks)) {
case Abort:
return fmt.Errorf("%s", combineDiskErrs(storageDisks, sErrs))
case FormatDisks:
console.Eraseline()
printFormatMsg(endpoints, storageDisks, printOnceFn())
return initFormatXL(storageDisks)
case InitObjectLayer:
console.Eraseline()
// Validate formats loaded before proceeding forward.
err := genericFormatCheckXL(formatConfigs, sErrs)
if err == nil {
printRegularMsg(endpoints, storageDisks, printOnceFn())
}
return err
case SuggestToHeal:
// Validate formats loaded before proceeding forward.
err := genericFormatCheckXL(formatConfigs, sErrs)
if err == nil {
printHealMsg(endpoints, storageDisks, printOnceFn())
}
return err
case WaitForQuorum:
console.Printf(
"Initializing data volume. Waiting for minimum %d servers to come online. (elapsed %s)\n",
len(storageDisks)/2+1, getElapsedTime(),
)
case WaitForConfig:
// Print configuration errors.
printConfigErrMsg(storageDisks, sErrs, printOnceFn())
case WaitForAll:
console.Printf("Initializing data volume for first time. Waiting for other servers to come online (elapsed %s)\n", getElapsedTime())
case WaitForFormatting:
console.Printf("Initializing data volume for first time. Waiting for first server to come online (elapsed %s)\n", getElapsedTime())
}
case <-globalServiceDoneCh:
return errors.New("Initializing data volumes gracefully stopped")
}
}
}
// Initialize storage disks based on input arguments.
func initStorageDisks(endpoints EndpointList) ([]StorageAPI, error) {
// Bootstrap disks.
storageDisks := make([]StorageAPI, len(endpoints))
for index, endpoint := range endpoints {
// Intentionally ignore disk not found errors. XL is designed
// to handle these errors internally.
storage, err := newStorageAPI(endpoint)
if err != nil && err != errDiskNotFound {
return nil, err
}
storageDisks[index] = storage
}
return storageDisks, nil
}
// Format disks before initialization of object layer.
func waitForFormatXLDisks(firstDisk bool, endpoints EndpointList, storageDisks []StorageAPI) (formattedDisks []StorageAPI, err error) {
if len(endpoints) == 0 {
return nil, errInvalidArgument
}
if storageDisks == nil {
return nil, errInvalidArgument
}
// Retryable disks before formatting, we need to have a larger
// retry window so that we wait enough amount of time before
// the disks come online.
retryDisks := make([]StorageAPI, len(storageDisks))
for i, storage := range storageDisks {
retryDisks[i] = &retryStorage{
remoteStorage: storage,
maxRetryAttempts: globalStorageInitRetryThreshold,
retryUnit: time.Second,
retryCap: time.Second * 30, // 30 seconds.
offlineTimestamp: UTCNow(),
}
}
// Start retry loop retrying until disks are formatted properly, until we have reached
// a conditional quorum of formatted disks.
err = retryFormattingXLDisks(firstDisk, endpoints, retryDisks)
if err != nil {
return nil, err
}
// Initialize the disk into a formatted disks wrapper.
formattedDisks = make([]StorageAPI, len(storageDisks))
for i, storage := range storageDisks {
// After formatting is done we need a smaller time
// window and lower retry value before formatting.
formattedDisks[i] = &retryStorage{
remoteStorage: storage,
maxRetryAttempts: globalStorageRetryThreshold,
retryUnit: time.Millisecond,
retryCap: time.Millisecond * 5, // 5 milliseconds.
offlineTimestamp: UTCNow(), // Set timestamp to prevent immediate marking as offline
}
}
// Success.
return formattedDisks, nil
}