I am writing a script that will be polling a directory looking for new files.
In this scenario, is it necessary to do some sort of error checking to make sure the files are completely written prior to accessing them?
I don't want to work with a file before it has been written completely to disk, but because the info I want from the file is near the beginning, it seems like it could be possible to pull the data I need without realizing the file isn't done being written.
Is that something I should worry about, or will the file be locked because the OS is writing to the hard drive?
This is on a Linux system.
Typically on Linux, unless you're using locking of some kind, two processes can quite happily have the same file open at once, even for writing. There are three ways of avoiding problems with this:
Locking
By having the writer apply a lock to the file, it is possible to prevent the reader from reading the file partially. However, most locks are advisory so it is still entirely possible to see partial results anyway. (Mandatory locks exist, but a strongly not recommended on the grounds that they're far too fragile.) It's relatively difficult to write correct locking code, and it is normal to delegate such tasks to a specialist library (i.e., to a database engine!) In particular, you don't want to use locking on networked filesystems; it's a source of colossal trouble when it works and can often go thoroughly wrong.
Convention
A file can instead be created in the same directory with another name that you don't automatically look for on the reading side (e.g., .foobar.txt.tmp) and then renamed atomically to the right name (e.g., foobar.txt) once the writing is done. This can work quite well, so long as you take care to deal with the possibility of previous runs failing to correctly write the file. If there should only ever be one writer at a time, this is fairly simple to implement.
Not Worrying About It
The most common type of file that is frequently written is a log file. These can be easily written in such a way that information is strictly only ever appended to the file, so any reader can safely look at the beginning of the file without having to worry about anything changing under its feet. This works very well in practice.
There's nothing special about Python in any of this. All programs running on Linux have the same issues.
On Unix, unless the writing application goes out of its way, the file won't be locked and you'll be able to read from it.
The reader will, of course, have to be prepared to deal with an incomplete file (bearing in mind that there may be I/O buffering happening on the writer's side).
If that's a non-starter, you'll have to think of some scheme to synchronize the writer and the reader, for example:
explicitly lock the file;
write the data to a temporary location and only move it into its final place when the file is complete (the move operation can be done atomically, provided both the source and the destination reside on the same file system).
If you have some control over the writing program, have it write the file somewhere else (like the /tmp directory) and then when it's done move it to the directory being watched.
If you don't have control of the program doing the writing (and by 'control' I mean 'edit the source code'), you probably won't be able to make it do file locking either, so that's probably out. In which case you'll likely need to know something about the file format to know when the writer is done. For instance, if the writer always writes "DONE" as the last four characters in the file, you could open the file, seek to the end, and read the last four characters.
Yes it will.
I prefer the "file naming convention" and renaming solution described by Donal.
Related
I saw a code where they were using file.flush(). so, i searched around and found This SO post. I kind of understood why there's a flush method. In the answer which was marked as answer the following was written
Typically you don't need to bother with either method, but if you're in a scenario where paranoia about what actually ends up on disk is a good thing, you should make both calls as instructed.
So, i was wondering, when we open a file using context manager and write some text, and then when the code exits from this context manager, there might be a chance that the text might not have been written to the file? if yes, why not python does this internally when file.close() is called? is it being done already?
The file objects in the io module (the ones you get from open) and everywhere else you'd expect in the stdlib always flush when they close, or rely on platform APIs that are guaranteed to do so.
Even third-party libraries are required to "close and flush the stream" on their close methods if they want their objects to be file objects.1
The main reason to call flush is when you're not closing the file yet, but some other program might want to see the contents.
For example, a lot of people write code like this:
with open('dump.txt', 'w') as f:
while True:
buf = read_off_some_thingy()
f.write(buf.decode())
time.sleep(5)
… and then they wonder why when they cat dump.txt or open it in Notepad or whatever, it's empty, or missing the last 3 lines, or cuts off in the middle of a line. That's the problem flush solves:
with open('dump.txt', 'w') as f:
while True:
buf = read_off_some_thingy()
f.write(buf.decode())
f.flush()
time.sleep(5)
Or, alternatively, they're running the same code, but the problem is that someone might pull the plug on the computer (or, more likely nowadays, kill your container), and then after restart they'll have a corrupt file that cuts off in mid-line and now the perl script that scans the output won't run and nobody wants to debug perl code. Different problem, same solution.
But if you know for a fact that the file is going to be closed by some point (say, because there's a with statement that ends before there), and you don't need the file to be done before that point, you don't need to call flush.
You didn't mention fsync, which is a whole other issue—and a whole lot more complicated than most people thing—so I won't get into it. But the question you linked already covers the basics.
1. There's always the chance that you're using some third-party library with a file-like object that duck-types close enough to a file object for your needs, but isn't one. And such a type might have a close that doesn't flush. But I honestly don't think I've ever seen an object that had a flush method, that didn't call it on close.
Python does flush the file when it's .close()d, which happens when the context manager is exited.
The linked post refers more to a scenario where you have, say, a log file that's open for a long time, and you want to ensure that everything gets written to disk after each write. That's where you'd want to .write(); .flush();.
I would like to check every minute if there was a file like "RESULTS.ODB" generated and if this file is bigger than 1.5 Gigabyte there starts another subprocess to get the Data from this file. How can i make sure that the file isn´t in progress to be written and everything is included?
I hope you know what i mean. Any ideas how to handle that?
Thank you very much. :)
If you have no control over the writing process, then you are at some point bound to fail somewhere.
If you do have control over the writer, a simple way to "lock" files is to create a symlink. If your symlink creation fails, there is already a write in progress. If it succeeds, you just acquired the "lock".
But if you do not have any control over writing and creation of the file, there will be trouble. You can try the approach as outlined here: Ensuring that my program is not doing a concurrent file write
This will read timestamps of the file and "guess" from them if writing has completed or not. This is more reliable than checking the file size, as you could end up with a file over your size threshold but writing still in progress.
In this case the problem would be the writer starting to write before you have read the file in its entirety. Now your reader would fail when the file it was reading disappeared half way through.
If you are on a Unix platform, you have no control over write and you absolutely need to do this, I would do something like this:
Check if file exists and if it does, if the "last written" timestamp
is "old enough" for me to assume the file is there
Rename the file to a different name
Check the renamed file that it still matches your criteria
Get data from the renamed file
Nevertheless, this will eventually fail and you will lose an update, as there is no way to make this atomic. Renaming will remove the problem of overwriting the file before you have read it, but if the writer decides to start writing between 1 and 2, you not only will receive an incomplete file but you might also break the writer if it does not like the file disappearing half way through.
I would rather try to find a way to somehow chain the actions together. Either your writer triggering the read process or adding a locking mechanism. Writing 1.5GB of data is not instantaneous and eventually the unexpected will happen.
Or if you definitely cannot do anything like that, could you ensure for example that your writer writes maximum once in N minutes or so? If you could be sure it never writes twice within a 5 minute window, you would wait in your reader until the file is 3 minutes old and then rename it and read the renamed file. You could also check if you could prevent the writer from overwriting. If you can do this, then you can safely process the file in your reader when it is "old enough" and has not changed in whatever grace period you decide to give it, and when you have read it, you will delete the file allowing the next update to appear.
Without knowing more about your environment and processes involved this is the best I can come up with. But there is no universal solution to this problem. It needs a workaround that is tailored to your particular environment.
In Python, if you either open a file without calling close(), or close the file but not using try-finally or the "with" statement, is this a problem? Or does it suffice as a coding practice to rely on the Python garbage-collection to close all files? For example, if one does this:
for line in open("filename"):
# ... do stuff ...
... is this a problem because the file can never be closed and an exception could occur that prevents it from being closed? Or will it definitely be closed at the conclusion of the for statement because the file goes out of scope?
In your example the file isn't guaranteed to be closed before the interpreter exits. In current versions of CPython the file will be closed at the end of the for loop because CPython uses reference counting as its primary garbage collection mechanism but that's an implementation detail, not a feature of the language. Other implementations of Python aren't guaranteed to work this way. For example IronPython, PyPy, and Jython don't use reference counting and therefore won't close the file at the end of the loop.
It's bad practice to rely on CPython's garbage collection implementation because it makes your code less portable. You might not have resource leaks if you use CPython, but if you ever switch to a Python implementation which doesn't use reference counting you'll need to go through all your code and make sure all your files are closed properly.
For your example use:
with open("filename") as f:
for line in f:
# ... do stuff ...
Some Pythons will close files automatically when they are no longer referenced, while others will not and it's up to the O/S to close files when the Python interpreter exits.
Even for the Pythons that will close files for you, the timing is not guaranteed: it could be immediately, or it could be seconds/minutes/hours/days later.
So, while you may not experience problems with the Python you are using, it is definitely not good practice to leave your files open. In fact, in cpython 3 you will now get warnings that the system had to close files for you if you didn't do it.
Moral: Clean up after yourself. :)
Although it is quite safe to use such construct in this particular case, there are some caveats for generalising such practice:
run can potentially run out of file descriptors, although unlikely, imagine hunting a bug like that
you may not be able to delete said file on some systems, e.g. win32
if you run anything other than CPython, you don't know when file is closed for you
if you open the file in write or read-write mode, you don't know when data is flushed
The file does get garbage collected, and hence closed. The GC determines when it gets closed, not you. Obviously, this is not a recommended practice because you might hit open file handle limit if you do not close files as soon as you finish using them. What if within that for loop of yours, you open more files and leave them lingering?
Hi It is very important to close your file descriptor in situation when you are going to use it's content in the same python script. I today itself realize after so long hecting debugging. The reason is content will be edited/removed/saved only after you close you file descriptor and changes are affected to file!
So suppose you have situation that you write content to a new file and then without closing fd you are using that file(not fd) in another shell command which reads its content. In this situation you will not get you contents for shell command as expected and if you try to debug you can't find the bug easily. you can also read more in my blog entry http://magnificentzps.blogspot.in/2014/04/importance-of-closing-file-descriptor.html
During the I/O process, data is buffered: this means that it is held in a temporary location before being written to the file.
Python doesn't flush the buffer—that is, write data to the file—until it's sure you're done writing. One way to do this is to close the file.
If you write to a file without closing, the data won't make it to the target file.
Python uses close() method to close the opened file. Once the file is closed, you cannot read/write data in that file again.
If you will try to access the same file again, it will raise ValueError since the file is already closed.
Python automatically closes the file, if the reference object has been assigned to some another file. Closing the file is a standard practice as it reduces the risk of being unwarrantedly modified.
One another way to solve this issue is.... with statement
If you open a file using with statement, a temporary variable gets reserved for use to access the file and it can only be accessed with the indented block. With statement itself calls the close() method after execution of indented code.
Syntax:
with open('file_name.text') as file:
#some code here
I've seen a few questions related to this but nothing that definitively answers my question.
I have a short python script that does some simple tasks, then outputs some text to a log file, waits for more input, and loops.
At times, the file is opened in write mode ("w") and other times it is opened in append mode ("a") depending on the results of the other tasks. For the sake of simplicity let's say it is in write mode/append mode 50/50.
I am opening files by saying:
with open(fileName, mode) as file:
and writing to them by saying:
file.write(line)
While these files are being opened, written to, appended to, etc., I expect a command prompt to be doing some read activities on them (findstr, specifically).
1) What's going to happen if my script tries to write to the same file the command window is reading from?
2) Is there a way to explicitly set the open to shared mode?
3)Does using the 'logger' module help at all/handle this instead of just manually making my own log files?
Thanks
What you are referring to is generally called a "race condition" where two programs are trying to read / write the same file at the same time. Some operating systems can help you avoid this by implementing a file-lock mutex system, but on most operating systems you just get a corrupted file, a crashed program, or both.
Here's an interesting article talking about how to avoid race conditions in python:
http://blog.gocept.com/2013/07/15/reliable-file-updates-with-python/
One suggestion that the author makes is to copy the file to a temp file, make your writes/appends there and then move the file back. Race conditions happen when files are kept open for a long time, this way you are never actually opening the main file in python, so the only point at which a collision could occur is during the OS copy / move operations, which are much faster.
I have a file that I want to read. The file may at any time be overwritten by another process. I do not want to block that writing. I am prepared to manage corruption to the data that I read, but do not want my reading to be in any way change the behaviour of the writing process.
The process that is writing the file is a delphi program running locally on the server. It opens the file using fmCreate. fmCreate tries to open the file exclusively and fails if there are any other handles on the file.
I am reading the file from a python script that accesses the file remotely across our network.
I am interested in whether there is a solution, independent of whether it is supported by python or delphi. I want to know if there is any way of achieving this under windows without modifying the writing program.
Edit: To reiterate, this is not a duplicate. The other question was trying to get read access to a file that is being written to. I want to the writer to have access to a file that I have open for reading. These are different questions (although I fear the answer will be similar, that it can't be done.)
I think the real answer here, all of these years later, is to use opportunistic locks. With this, you can open the file for read access, while telling the OS that you want to be notified if another program wants to access the file. Basically, you can use the file as long as you like, and then back off if someone else needs it. This avoids the sharing/access violation that the other program would normally get, if you had just opened the file "normally".
There is an MSDN article on Opportunistic Locks. Raymond Chen also has a blog article about this, complete with sample code: Using opportunistic locks to get out of the way if somebody wants the file
The key is calling the DeviceIoControl function, with the FSCTL_REQUEST_OPLOCK flag, and passing it the handle to an event that you previously created by calling CreateEvent.
It should be straightforward to use this from Delphi, since it supports calling Windows API functions. I am not so sure about Python. But, given the arrangement in the question, it should not be necessary to modify the Python code. Just make your Delphi code use the opportunistic lock when it opens the file, and let it get out of the way when the Python script needs the file.
Also much easier and lighter weight than a filter driver or the Volume Shadow Copy service.
You can setup a filter driver which can act in two ways: (1) modify the flags when the file is opened, and (2) it can capture the data when it's written to the file and save a copy of the data elsewhere.
This approach is much more lightweight and efficient than volume shadow copy service, mentioned in comments, however it requires having a filter driver. There exist several drivers on the market (i.e. those are products which include a driver and let you write business logic in user mode) yet they are costly and can be an overkill in your case. Still, if you need the thing for private use only, contact me privately for a license for our CallbackFilter.
Update: if you want to let the writer open the file which has been already opened, then a filter which will modify flags when the file is being opened is your only option.