JPH0282325A - Restoring system for event log data - Google Patents

Abstract

PURPOSE:To give a fault tolerance by providing a non-volatile memory or the like where each event data is written and restoring event control data and event log data from the non-volatile memory at the time of the occurrence of crush. CONSTITUTION:Each of events A to D is temporarily written in a buffer 2 at each time of the occurrence of the event, and events A to D are collectively outputted to an event log file 3 when the buffer 2 is filled up, and each of events E and F is written also when it occurs. Event data A to F are written in the event log file 3 simultaneously with write to the buffer 2. When crush occurs before or during the write of the event log to the event log file 2, event data A to F are restored from a non-volatile memory 4 to the buffer 2 because there is probability that even data from the buffer or the like is lost. Thus, the non-volatile memory 4 is used to give a fault tolerance to event log data to the event log file 3.

Description

[Detailed description of the invention] [Summary] Obstacle light 4
: Regarding the event log data restoration method that has a non-volatile memory that stores the log data necessary to restore the previous state, another event control is prepared in the standby space in addition to the currently operating event control. , the control information necessary for event control and the event log data to be written to the event log file are stored in non-volatile memory, and when a crash occurs, the event control data and event log data are recovered from the non-volatile memory to make it fault tolerant. The system is equipped with a buffer for writing each event data that occurs in the system, an event log file for actually updating the event data from the buffer, and a non-volatile memory for writing each of the event data. If a crash occurs before or during an update, the contents of the buffer are recovered from the nonvolatile memory and the event log file is actually updated.

[Industrial application field]

The present invention relates to a computer system, and more particularly, the present invention relates to an event log data restoration method that has a nonvolatile memory that stores log data necessary for restoring the computer system to the state immediately before the failure occurs in response to a failure occurring in the computer system. Regarding.

With the development of computer systems, there has been an increasing need to construct fault-tolerant computers that have the ability to continue operating without stopping even if a failure occurs.

In general, a chronological memory of changes made to the contents of a data set is called a log, and in particular, logs contain information necessary to restore the state immediately before the failure occurs in a computer system. This is called log data. To perform batch processing of log data acquisition, log data generated during a transaction is accumulated until the end of the transaction.

Therefore, events that occur in the system are managed by event control as an operating system, and when an event occurs, this event control outputs event data to an event log file that manages information on the occurrence, billing, etc.

[Conventional technology]

FIG. 4 is a conceptual diagram of a writing processing method to an event log file in a conventional system.

In l, event log files are log data, data A, data B, ... data Z, necessary to restore the state immediately before the failure occurs in the computer system when a failure occurs.
Store. In the conventional system, in order to accumulate log data generated during a transaction until the end of the transaction, log data for each event A to Z is written to the event log file 1 via a log buffer.

[Problem to be solved by the invention]

Therefore, in this type of conventional method, the event control itself or the cluster may crash during the writing process to the event log file 1 due to a CPU failure or doing something that is prohibited by software. If there is, the process will be interrupted midway. Therefore, the contents of the event data at the time of the crash may not be stored in the event log file, so not only will the contents not be compensated for, but the event log data may be erased.
Alternatively, there was a real problem that there was a risk that the event log itself could be destroyed.

In addition to the currently operating event control, another event control is prepared in the standby space, and control information necessary for the event control and event log data to be written to the event log file are stored in a nonvolatile memory. The purpose of this system is to restore event control data and event log data from non-volatile memory in the event of a crash, making it fault tolerant.

[Means to solve the problem]

FIG. 1 is a diagram illustrating the principle of the present invention and shows the contents of an event log file created by the fault tolerant system of the present invention.

The buffer 2 temporarily writes the events A to D each time they occur, and outputs the events A to D all at once to the event log file 3 when it becomes full.

Write again when the next event E-F occurs.

Event log file 3 stores each event data A-F in buffer 2.
Write at the same time as writing to. If a crash occurs before or during the writing of the event log to the event log file 3, the event data may disappear from the buffer, etc., so event data A to F are transferred from the non-volatile memory 4 to the buffer 2. Recover.

[For production]

In the present invention, a standby space is prepared in addition to the space operated by event control, and when the event control itself crashes, continuous movement is possible by switching control to the standby space. Event log data generated until the actual update is performed is stored in non-volatile memory 4.
The content of the non-volatile memory 4 is erased when the event log file 3 is actually updated.If a crash occurs and the data is restored, the event data is stored in the non-volatile memory 4. The system recovers the data and writes the data to the event log data 3 again.

〔Example〕

Next, embodiments of the present invention will be described with reference to the drawings.

FIG. 2 is a conceptual diagram of the event log data restoration processing method of the present invention. In the same figure, reference numeral 11 denotes an event control means consisting of an operating system that controls events occurring in the system, and reference numeral 12 denotes a first event control means provided in a standby space that controls events when the event control means 11 itself crashes. a second event control means consisting of an operating system similar to the event control means 11; 13 a buffer for speeding up log data output processing; 14 a buffer for storing log data related to failures occurring within the computer system; 15 is a non-volatile memory that temporarily stores event log data related to events occurring in the system until the event log file 14 is actually updated; 16 is an event control means 12 in the standby space; is a buffer controlled by

In the present invention, events occurring in the system are managed and operated by the first event control means 11, but in addition to the virtual space in which this operation takes place, another standby space is prepared that coexists with a part of the virtual space. However, when the first event control means 11 itself crashes, control is switched to the second event control means 12 located in the standby space, so that even if a failure occurs in the system, the operation continues without stopping. It has functions. In order to perform this function, whenever an event occurs, event log data is written to the buffer 13 and at the same time, the event log data is stored in the non-volatile memory 15, that is, a fault-tolerant memory (FTM).
When event data is actually updated to the event log file 14 via the buffer 13, the same event log data in the nonvolatile memory 15 is erased from the nonvolatile memory 15. For example, when the power is cut off, the first event control means 11
Even if the event control means 11 itself or the cluster crashes while writing to the event log file 14, the log data at that point is stored in the non-volatile memory 15, so when the crash is restored, the log data is stored in the non-volatile memory 15. Control is switched to the second event control means 2, and event data is restored from the nonvolatile memory to the buffer 16, and the data is written to the event log file 14 again.

The first event control means 11 includes a procedural process for scheduling writing to event log data, and when there is no crash, the log data writing process is first buffered in the order of event data 1, 2, and 3, for example. , will not be written to the event log file until buffer 13 is full. This log write process is performed using write macro command 1.
It will be held at 7. That is, when all events occur, a request is made to output them to the event log file 14, but rather than writing the data directly to the event log file 14, one day huffering is performed to reduce the number of times.

This buffering is for processing the output of data and writing it to the buffer 13, which has a relatively fast processing time.If it is buffered, it returns by just writing to the buffer 13, and writes it to the event log file 14. will not be performed. Buffer 13 by writing to buffer 13
By writing to the event log file 14 at once when it becomes full, processing speed is increased.

The nonvolatile memory 15 is such that even if the power is turned off, the contents of the memory remain without being erased, and the written contents can be outputted without being destroyed upon recovery.

In the first event control means 11, buffering is performed by the write macro instruction 17, but when the buffering is finished and the buffer 13 is full, the event data 1 in the buffer is changed to the event data ■, that is, all of the buffer 13 is The contents are written to the event log file 14.

When this writing is completed, the event log file 14 is actually updated. The first event control means 11 determines whether or not this actual update has occurred (18)L, and the event log data written to the event log file 14 in the actual update is erased from the nonvolatile memory 15 (19).

If there is no actual update, that is, if a buffering operation is currently in progress, the contents written to the nonvolatile memory 15 will not be erased and will remain as they are.

If a crash occurs in the space of the first event control means 11, it is necessary to compensate for the contents of the event data at the time of the crash. Especially event log file 14
If the event control means 11 itself or the cluster crashes during the writing process, conventionally, the contents of the event log data at the time of the crash are not saved anywhere and may not be compensated for. However, in the present invention, since the event log data is always buffered in the non-volatile memory 15 during normal times and the event log data written in the event log file 14 is also written in the non-volatile memory 15, there is no risk that the event log data will be deleted and a crash will occur. The event log data is recovered from the nonvolatile memory 15 at the time when the event log data is recovered from the nonvolatile memory 15, and the log data is written again to the event log file 14 via the buffer 16 under the control of the event control means 12. Note that the event data to be recovered when a crash occurs is the same data as the content written to the buffer, and is the content from event data 1 to event data ■.Therefore, on the event log file 14, the event data that is being written to the event log file 14 is being processed. Even in the event of a crash, event data that has already been written to the event log file 14 may be subject to double writing processing. However, this poses no risk of erasing event log data or corrupting event log files.

Also, in Figure 2, the buffer (1
3, 16) exists in a form separate from the non-volatile memory 15, but if the non-volatile memory 15 is quite fast, it is the same as the buffer that speeds up and the non-volatile memory that compensates for event log data in the event of a crash. You can also use it as

Further, the event control means 12 in the waiting space is the event control means 1
This is common to the control content of 1, and the event control means 11
Even if the event control means 11 crashes, the control data necessary to perform the same operation as the event control means 11 is stored in the non-volatile memory 15. Non-volatile memory 15
By transferring control to the event control means 12 in the standby space via the event control means 12, log writing processing at the time of a crash is executed.

FIG. 3 is a processing flow diagram of the event log data restoration method of the present invention. When a notification for event log data writing processing is issued from another control means, the event control means 11
Execute an event log data output command in the TI. When this output command is decoded, event log data is stored in buffer 1.
3 (Sr2), and the same event log data is also stored in the nonvolatile memory 15 (Sr2). Then, a linked return is performed to notify other controls that buffering has occurred. At Sr1, it is detected whether the buffer is full with respect to the written event data, and if it is full, the contents of the buffer are actually updated to the event log file 14 (Sr1). If the buffer is not full, only the contents of the buffer are left as is, and actual updates to the event log file are not written from the buffer to the event log file until the buffer is full.

Writing processing of such event log data (ST1-5
T4) is performed by the first event control means 11.

In such a normal state, if a failure or crash occurs on the system, especially if the event control means 11 itself or the cluster crashes during the writing process to the event log file 14, the present invention enables continuous operation. The second event control means 1 on the standby space controls the
I will move to 2. Since the same event log data is also stored in the nonvolatile memory 5 during buffering, if a crash occurs, the event data is recovered from the nonvolatile memory 15 (Sr5). Then, the recovered log data is actually updated to the event log file 14 via the buffer 16 of the second event control means 12. Sr6 after actual update
At this point, the contents of the buffer and the contents of non-volatile memory are cleared and the normal processing state is restored. Control from the time of crash occurrence to the time of crash recovery (Sr5.
3T6) is performed by the second event control means 12 located on the standby space.

In this way, in the present invention, even if a crash occurs and the buffer is destroyed before the actual update to the event log file is completed, event control is switched to the event control means 12 in the standby space, so that the event can be controlled. It is possible to treat the occurrence without losing it.

〔Effect of the invention〕

In the present invention, it is possible to make event log data in an event log file fault tolerant using nonvolatile memory, and it is possible to guarantee the contents of each event log data in the event log file and to prevent file destruction. It becomes possible to view the image. Therefore, it is possible to provide a computer system that continues to operate without stopping even if a failure occurs.

[Brief explanation of the drawing]

Fig. 1 is a functional diagram of the principle of the present invention. Fig. 2 is a conceptual diagram of the event log data restoration processing method of the invention. Fig. 3 is a processing flow diagram of the event log data restoration method of the invention. Fig. 4 1 is a conceptual diagram of a writing processing method to an event log file in a conventional system. 2...Buffer, 3...Event log file, 4...Non-volatile memory. Patent Applicant: Fujitsu Ltd. Rinri Kyouryuzu Diagram Major 2 Occurrence...Log Pressure Elephant Log Aisle 3 Resume Matasato Method Diagram

Claims (1)

[Claims] 1) A buffer (2) for writing each event data generated in the system; an event log file (3) for actually updating the event data from the buffer (2); and a non-volatile memory (4) for writing, and if a crash occurs before or during the actual update of the event data, the contents of the buffer (2) are recovered from the non-volatile memory (4) and the event log file is restored. (3) A method for restoring event log data characterized by performing an actual update. 2) A first event control means (11) that controls and further manages events occurring within the system, and is placed on another virtual space that is shared with or different from the virtual space in which the first event control means operates. A second event control means (12) is provided on the standby space to perform subsequent control when the first event control means (11) itself crashes, and a second event control means (12) that performs subsequent control when the first event control means (11) itself crashes; Event log file means (14) for storing log data necessary for restoring to the state immediately before the occurrence, and an event that is managed by the first event control means (11) when an event occurs in the system. and nonvolatile memory means (15) for storing event log data before storing the event log data that has occurred in the event log file means (14),
the second event control means (12) when restoring from a failure;
Restoration of event log data characterized by reading out at least event log data from the non-volatile memory means (15) under control of the controller, writing the event log data again into the event log file means (14), and restoring the state. method. 3) The non-volatile memory means (15) exists separately from the buffer means (13, 16) for speeding up the writing operation of event log data to the event log file, and the event data generated in the system is The buffer means (13)
At the same time, the event data is written to the event log file via the buffer means (16) and also stored in the non-volatile memory (15), and upon recovery from a crash, the event data is transferred from the non-volatile memory (15) to the event log file via the buffer means (16). 3. The method for restoring event log data according to claim 2, wherein the event log data is rewritten to the means (14). 4) The event log data restoration method according to claim 3, characterized in that said nonvolatile memory means (15) and buffer means (13, 16) are also used. 5) The event log data restoration method according to claim 2, wherein the contents of the nonvolatile memory means (15) are cleared after restoring the event data from the nonvolatile memory means (15).