Before changes are made to an Exchange database file, Exchange writes the changes to a transaction log file. After a change has been safely logged, it can then be written to the database file. It is common for these changes to become available to end users just after the changes have been secured to the transaction log, but before they have been written to the database file.
Exchange employs a sophisticated internal memory management system that is tuned for high performance and can efficiently manage the caching of dozens of gigabytes (GBs) of database pages. Therefore, physically writing out changes to the database file is a low-priority task during normal operation.
If a database suddenly stops, cached changes are not lost just because the memory cache was destroyed. When the database restarts, Exchange scans the log files, and reconstructs and applies any changes not yet written to the database file. This process is called replaying log files. The database is structured so that Exchange can determine whether any operation in any log file has already been applied to the database, needs to be applied to the database, or does not belong to the database.
Rather than write all log information to a single large file, Exchange uses a series of log files, each exactly one megabyte, or 1,024 kilobytes (KB), in size. When a log file is full, Exchange closes it and renames it with a sequential number. The first log that is filled ends with the name Enn00000001.log. The nn refers to a two-digit number known as the base name or log prefix.
Log files for each database are distinguished by file names with numbered prefixes (for example, E00, E01, E02, or E03). The log file currently open for a database is simply named Enn.log—it does not have a sequence number until it has been filled and closed.
The checkpoint file (Enn.chk) tracks how far Exchange has progressed in writing logged information to the database files. There is a checkpoint file for each log stream, and a separate log stream for each database. All the databases share a single log stream. Thus, a single log file often contains operations for multiple databases.
Log files are numbered in a hexadecimal manner, so the log file after E0000000009.log is E000000000A.log, not E0000000010.log. You can convert log file sequence numbers to their decimal values by using the Windows Calculator (Calc.exe) application in Scientific mode. To do this, run Calc.exe, and then, from the View menu, click Scientific.
To view the decimal sequence number for a specific log file, you can examine its header by using the Exchange Server Database Utilities (Eseutil.exe) tool. The first 4-KB page of each log file contains header information that describes and identifies the log file and the databases it belongs to. The command Eseutil /ml [log file name] displays the header information.
If you use the wrong switch for displaying a header (for example, by using /ml with a database header instead of /mh), an error is displayed or the header information that is displayed may be garbled or incorrect.
You cannot view the header of a database while it is mounted. You also cannot view the header of the current log file (Enn.log) while any database is mounted. Exchange holds the current log file open as long as one database is using it. You can, however, view the checkpoint file header while databases are mounted. Exchange updates the checkpoint file every thirty seconds, and its header is viewable except during the moment when an update is occurring.
As an Exchange administrator, it is valuable to understand Exchange file headers. If you understand the file headers, you can determine which database and log files belong together and which files are needed for successful recovery.
In the following log file header example, note the first four lines.
Base name: e00
Log file: e00.log
lGeneration: 11 (0xB)
Checkpoint: (0xB,7DC,6F)
These log file header lines show that this log file is the current log file because the log file name does not have a sequence number. The lGeneration line shows that when the log is filled and closed, its sequence number will be B, corresponding to the decimal value 11. The base name is e00, and therefore the final log file name will be E000000000B.log.
The Checkpoint value in the previous header example is not actually read from the log file header, but it is displayed as if it were. Eseutil.exe reads the Checkpoint value directly from Enn.chk, so you do not have to enter a separate command to learn where the checkpoint file is. If the checkpoint file has been destroyed, the Checkpoint value reads NOT AVAILABLE. In this case, the checkpoint is in the current log file (0xB), and the numbers 7DC and 6F indicate how far into the log file the checkpoint is. Note that you will seldom have a practical need for this information.
If the checkpoint file is destroyed, Exchange can still recover and replay log files appropriately. But to do so, Exchange begins scanning log files, beginning with the oldest file available, instead of starting at the checkpoint log. Exchange skips data that has already been applied to the database and works sequentially through the logs until data that must be applied is encountered.
Typically, it takes only one or two seconds for Exchange to scan a log file that has already been applied to the database. If there are operations in a log file that must be written to the database, it can take anywhere from 10 seconds to several minutes to apply them. On average, a log file's contents can be written to the database in 30 seconds or less.
When an Exchange database shuts down normally, all outstanding data is written to the database files. After normal shutdown, the database file set is considered consistent, and Exchange detaches it from its log stream. This means that the database files are now self-contained—they are completely up to date. The transaction logs are not required to start the database files.
You can tell whether a database has been shut down cleanly by running the command Eseutil /mh and examining the file headers.
With all databases disconnected and in a Clean Shutdown state, all log files can be safely deleted without affecting the databases. If you were then to delete all log files, Exchange would generate a new sequence of logs starting with Enn00000001.log. You could even move the database files to a different server that has existing log files, and the databases would attach themselves to a different log stream.
.gif) Note: |
|---|
|
Although you can delete the log files after all databases have been shut down, doing so will affect your ability to restore older backups and roll forward. The current database no longer needs the existing log files, but they may be necessary if you must restore an older database.
|
If a database is in a Dirty Shutdown state, all existing transaction logs from the checkpoint forward must be present before you can mount the database again. If these logs are unavailable, you must repair the database by running the command Eseutil /p to make the database consistent and ready to start.
.gif) Caution: |
|---|
|
If you have to repair a database, some data will be lost. Data loss is frequently minimal; however, it may be catastrophic. After running Eseutil /p on a database, you should completely repair the database with the following two operations: First, run Eseutil/d to defragment the database. This operation discards and rebuilds all database indexes and space trees.
|
In addition to allowing Exchange to recover reliably from an unexpected database stop, transaction logging is also essential to making and restoring online backups. For more information about making and restoring online backups, see Understanding Backup, Restore and Disaster Recovery.