Re: How to safely maintain a status file
Am 13.07.2012 03:52, schrieb Steven D'Aprano: And some storage devices (e.g. hard drives, USB sticks) don't actually write data permanently even when you sync the device. They just write to a temporary cache, then report that they are done (liar liar pants on fire). Only when the cache is full, or at some random time at the device's choosing, do they actually write data to the physical media. The result of this is that even when the device tells you that the data is synched, it may not be. Yes, that's another issue. Either you have to buy expensive enterprise hardware with UPS batteries or you need to compensate for failures on software level (e.g. Hadoop cluster). We have big storage devices with double redundant controllers, on board buffer batteries, triple redundant power supplies, special RAID disks, multipath IO fiber channel links and external backup solution to keep our data reasonable safe. Christian -- http://mail.python.org/mailman/listinfo/python-list
Re: How to safely maintain a status file
On Fri, Jul 13, 2012 at 2:26 PM, rantingrickjohn...@gmail.com wrote: On Thursday, July 12, 2012 10:13:47 PM UTC-5, Steven D#39;Aprano wrote: Rick has obviously never tried to open a file for reading when somebody else has it opened, also for reading, and discovered that despite Windows being allegedly a multi-user operating system, you can#39;t actually have multiple users read the same files at the same time. You misread my response. My comment was direct result of Christian stating: (paraphrase) On some systems you are not permitted to delete a file whilst the file is open ...which seems to be consistent to me. Why would *anybody* want to delete a file whilst the file is open? POSIX doesn't let you delete files. It lets you dispose of filenames. Python does the same with its 'del'. The object (file) exists until the system decides otherwise. Here's a simpler example: Hardlinks. Suppose you have two names pointing to the same file; are you allowed to unlink one of them while you have the other open? ChrisA -- http://mail.python.org/mailman/listinfo/python-list
Re: How to safely maintain a status file
On Thu, 12 Jul 2012 21:26:20 -0700, rantingrickjohnson wrote: On Thursday, July 12, 2012 10:13:47 PM UTC-5, Steven D#39;Aprano wrote: Rick has obviously never tried to open a file for reading when somebody else has it opened, also for reading, and discovered that despite Windows being allegedly a multi-user operating system, you can#39;t actually have multiple users read the same files at the same time. You misread my response. My comment was direct result of Christian stating: (paraphrase) On some systems you are not permitted to delete a file whilst the file is open ...which seems to be consistent to me. Why would *anybody* want to delete a file whilst the file is open? Because it is useful and a sensible thing to do. Why should one misbehaved application, keeping a file open, be allowed to hold every other application, and the file system, hostage? This is one of the many poor decisions which makes Windows so vulnerable to viruses and malware. If malware can arrange to keep itself open, you can't delete it. Thanks guys! Bringing back the car analogy again: Would you consider jumping from a moving vehicle a consistent interaction with the interface of a vehicle? Of course not. The interface for a vehicle is simple and consistent: 1. You enter the vehicle at location A 2. The vehicle transports you to location B 3. You exit the vehicle Amusingly, you neglected to specify the vehicle stops -- and rightly so, because of course having to stop the vehicle is not a *necessary* condition for exiting it, as tens of thousands of stunt men and women can attest. Not to mention people parachuting out of an airplane, pirates or commandos boarding a moving ship, pedestrians transferring from a slow moving walkway to a faster moving walkway, farmers jumping off a trailer while it is still being towed behind a tractor (and jumping back on again), and Bruce Willis in Red in very possibly the best slow-motion action sequence in the history of Hollywood. http://www.youtube.com/watch?v=xonMpj2YyDU At no time during the trip would anyone expect you to leap from the vehicle. Expected or not, you can do so. But when you delete open files, you are essentially leaping from the moving vehicle! This behavior goes against all expectations of consistency in an API -- and against all sanity when riding in a vehicle! Fortunately, files on a file system are not cars, and deleting open files is a perfectly reasonable thing to do, no more frightening than in Python deleting a reference to an object using the del statement. Imagine how stupid it would be if this happened: py x = 42 py y = x py del y Traceback (most recent call last): File stdin, line 1, in module DeleteError: cannot delete reference to object '42' until no other references to it exist Fortunately, Python doesn't do that -- it tracks when the object is no longer being accessed, and only then physically reclaims the memory used. And so it is on POSIX file systems: the file system keeps track of when the file on disk is no longer being accessed, and only then physically reclaims the blocks being used. Until then, deleting the file merely unlinks the file name from the blocks on disk, in the same way that del y merely unlinks the name y from the object 42. Opening files for exclusive read *by default* is a pointless and silly limitation. It#39;s also unsafe: if a process opens a file for exclusive read, and then dies, *no other process* can close that file. Oh come on. Are you actually going to use errors or unintended consequences, or even Acts of God to defend your argument? Features have to be judged by their actual consequences, not some unrealistic sense of theoretical purity. The actual consequences of mandatory exclusive file locking is, *it sucks*. Windows users are used to having to reboot their server every few days because something is broken, so they might not mind rebooting it because some file is locked in a mandatory open state and not even the operating system can unlock it. But for those with proper operating systems who expect months of uninterrupted service, mandatory locking is a problem to be avoided, not a feature. Okay. Okay. I suppose IF the car spontaneously combusted THEN the passengers would be wise to jump out, leaving the vehicle to the whims of inertia. In this analogy, is the car the file name, the inode, or the directory? Are the passengers the file name(s), or the file contents, or the inode? Is the driver meant to be the file system? If I have a hard link to the file, does that mean the passengers are in two cars at once, or two lots of passengers in the same car? One neat trick is to open a file, then delete it from disk while it is still open. So long as your process is still running, you can write to this ghost file, as normal, but no other process can (easily) see it. And when your process ends, the file contents is
RE: How to safely maintain a status file
Well neat tricks aside, I am of the firm belief that deleting files should never be possible whilst they are open. This is one of the few instances I think Windows does something better than OS X. Windows will check before you attempt to delete (i.e. move to Recycling Bin) while OS X will move a file to Trash quite happily only tell me it cannot remove the file when I try to empty the Trash. Ramit Ramit Prasad | JPMorgan Chase Investment Bank | Currencies Technology 712 Main Street | Houston, TX 77002 work phone: 713 - 216 - 5423 -- This email is confidential and subject to important disclaimers and conditions including on offers for the purchase or sale of securities, accuracy and completeness of information, viruses, confidentiality, legal privilege, and legal entity disclaimers, available at http://www.jpmorgan.com/pages/disclosures/email. -- http://mail.python.org/mailman/listinfo/python-list
Re: [Python] RE: How to safely maintain a status file
On 07/13/2012 11:00 AM, Prasad, Ramit wrote: Well neat tricks aside, I am of the firm belief that deleting files should never be possible whilst they are open. This is one of the few instances I think Windows does something better than OS X. Windows will check before you attempt to delete (i.e. move to Recycling Bin) while OS X will move a file to Trash quite happily only tell me it cannot remove the file when I try to empty the Trash. While I was trained in the Unix way, and believe it is entirely appropriate to delete an open file. Even if I my program is the opener. It's just too handy to have temp files that disappear on their own. As opposed to periodically going to %TEMP% and deleting them manually. Gah. -- Chris. -- http://mail.python.org/mailman/listinfo/python-list
RE: [Python] RE: How to safely maintain a status file
Well neat tricks aside, I am of the firm belief that deleting files should never be possible whilst they are open. This is one of the few instances I think Windows does something better than OS X. Windows will check before you attempt to delete (i.e. move to Recycling Bin) while OS X will move a file to Trash quite happily only tell me it cannot remove the file when I try to empty the Trash. While I was trained in the Unix way, and believe it is entirely appropriate to delete an open file. Even if I my program is the opener. It's just too handy to have temp files that disappear on their own. As opposed to periodically going to %TEMP% and deleting them manually. Gah. In my experience things that are too handy are usually breaking what I consider right. That being said, I am not entirely sure what I think is right in this circumstance. I suppose it depends on if I am the person deleting or the person who is looking at a file that is being deleted. Or the user who just wants the stupid computer to just Work. I lean slightly towards the POSIX handling with the addition that any additional write should throw an error. You are now saving to a file that will not exist the moment you close it and that is probably not expected. Ramit This email is confidential and subject to important disclaimers and conditions including on offers for the purchase or sale of securities, accuracy and completeness of information, viruses, confidentiality, legal privilege, and legal entity disclaimers, available at http://www.jpmorgan.com/pages/disclosures/email. -- http://mail.python.org/mailman/listinfo/python-list
Re: [Python] RE: How to safely maintain a status file
On Sat, Jul 14, 2012 at 3:59 AM, Prasad, Ramit ramit.pra...@jpmorgan.com wrote: I lean slightly towards the POSIX handling with the addition that any additional write should throw an error. You are now saving to a file that will not exist the moment you close it and that is probably not expected. There are several different possible right behaviors here, but they depend more on the application than anything else. With a log file, for instance, the act of deleting it is more a matter of truncating it (dispose of the old history), so the right thing to do is to start a fresh file. Solution: Close the file and re-open it periodically. But I don't know of an efficient way to do that with Windows semantics. Renaming/moving an open file in order to perform log rotation isn't all that easy. ChrisA -- http://mail.python.org/mailman/listinfo/python-list
Re: [Python] RE: How to safely maintain a status file
On 13/07/12 19:59:59, Prasad, Ramit wrote: I lean slightly towards the POSIX handling with the addition that any additional write should throw an error. You are now saving to a file that will not exist the moment you close it and that is probably not expected. I'd say: it depends. If the amount of data your script needs to process does not fit in RAM, then you may want to write some of it to a temporary file. On a Posix system, it's entirely normal to unlink() a temp file first thing after you've created it. The expectation is that the file will continue to exists, and be writeable, until you close it. In fact, there's a function in the standard library named tempfile.TemporaryFile that does exactly that: create a file and unlink it immediately. This function would be useless if you couldn't write to your temporary file. Hope this helps, -- HansM -- http://mail.python.org/mailman/listinfo/python-list
Re: [Python] RE: How to safely maintain a status file
On 13/07/2012 19:28, Hans Mulder wrote: On 13/07/12 19:59:59, Prasad, Ramit wrote: I lean slightly towards the POSIX handling with the addition that any additional write should throw an error. You are now saving to a file that will not exist the moment you close it and that is probably not expected. Strictly speaking, the file does exist, it's just that there are no names referring to it. When any handles to it are also closed, the file _can_ truly be deleted. As has been said before, in the *nix world, unlink _doesn't_ delete a file, it deletes a name. I'd say: it depends. If the amount of data your script needs to process does not fit in RAM, then you may want to write some of it to a temporary file. On a Posix system, it's entirely normal to unlink() a temp file first thing after you've created it. The expectation is that the file will continue to exists, and be writeable, until you close it. In fact, there's a function in the standard library named tempfile.TemporaryFile that does exactly that: create a file and unlink it immediately. This function would be useless if you couldn't write to your temporary file. It's possible to create a temporary file even in Windows. -- http://mail.python.org/mailman/listinfo/python-list
RE: How to safely maintain a status file
On 07/13/2012 12:59 PM, Prasad, Ramit wrote: I lean slightly towards the POSIX handling with the addition that any additional write should throw an error. You are now saving to a file that will not exist the moment you close it and that is probably not expected. Ramit But if I created, then deleted it while holding an open file descriptor, it is entirely likely that I intend to write to it. I'll admit, these days there are those in the Unix/Linux community that consider using an anonymous file a bad idea; I'm just not one of them. -- Chris. -- http://mail.python.org/mailman/listinfo/python-list
Re: [Python] RE: How to safely maintain a status file
Am 13.07.2012 21:57, schrieb MRAB: It's possible to create a temporary file even in Windows. Windows has a open() flag named O_TEMPORARY for temporary files. With O_TEMPORARY the file is removed from disk as soon as the file handle is closed. On POSIX OS it's common practice to unlink temporary files immediately after the open() call. -- http://mail.python.org/mailman/listinfo/python-list
Re: How to safely maintain a status file
On Fri, 13 Jul 2012 15:15:13 -0500, Chris Gonnerman wrote: On 07/13/2012 12:59 PM, Prasad, Ramit wrote: I lean slightly towards the POSIX handling with the addition that any additional write should throw an error. You are now saving to a file that will not exist the moment you close it and that is probably not expected. Ramit But if I created, then deleted it while holding an open file descriptor, it is entirely likely that I intend to write to it. I'll admit, these days there are those in the Unix/Linux community that consider using an anonymous file a bad idea; I'm just not one of them. A badly-behaved application can write oodles and oodles of data to an unlinked file, which has the result of temporarily using up disk space that doesn't show up when you do an ls. For an inexperienced system administrator, this may appear mysterious. The solution is to us lsof to identify the unlinked file, which gives you the process id of the application, which you can then kill. As soon as you do that, the space is freed up again. Like all powerful tools, unlinked files can be abused. Underpowered tools can't be abused, but nor can they be used. -- Steven -- http://mail.python.org/mailman/listinfo/python-list
Re: How to safely maintain a status file
You are contradicting yourself. Either the OS is providing a fully atomic rename or it doesn't. All POSIX compatible OS provide an atomic rename functionality that renames the file atomically or fails without loosing the target side. On POSIX OS it doesn't matter if the target exists. This is not a contradiction. Although the rename operation is atomic, the whole change status process is not. It is because there are two operations: #1 delete old status file and #2. rename the new status file. And because there are two operations, there is still a race condition. I see no contradiction here. You don't need locks or any other fancy stuff. You just need to make sure that you flush the data and metadata correctly to the disk and force a re-write of the directory inode, too. It's a standard pattern on POSIX platforms and well documented in e.g. the maildir RFC. It is not entirely true. We are talking about two processes. One is reading a file, another one is writting it. They can run at the same time, so flushing disk cache forcedly won't help. -- http://mail.python.org/mailman/listinfo/python-list
Re: How to safely maintain a status file
Renaming files is the wrong way to synchronize a crawler. Use a database that has ACID properties, such as SQLite. Far fewer I/O operations are required for small updates. It's not the 1980s any more. I agree with this approach. However, the OP specifically asked about how to update status file. -- http://mail.python.org/mailman/listinfo/python-list
Re: How to safely maintain a status file
Am 12.07.2012 14:30, schrieb Laszlo Nagy: This is not a contradiction. Although the rename operation is atomic, the whole change status process is not. It is because there are two operations: #1 delete old status file and #2. rename the new status file. And because there are two operations, there is still a race condition. I see no contradiction here. Sorry, but you are wrong. It's just one operation that boils down to point name to a different inode. After the rename op the file name either points to a different inode or still to the old name in case of an error. The OS guarantees that all processes either see the first or second state (in other words: atomic). POSIX has no operation that actually deletes a file. It just has an unlink() syscall that removes an associated name from an inode. As soon as an inode has no names and is not references by a file descriptor, the file content and inode is removed by the operating system. rename() is more like a link() followed by an unlink() wrapped in a system wide global lock. It is not entirely true. We are talking about two processes. One is reading a file, another one is writting it. They can run at the same time, so flushing disk cache forcedly won't help. You need to flush the data to disk as well as the metadata of the file and its directory in order to survive a system crash. The close() syscall already makes sure that all data is flushed into the IO layer of the operating system. With POSIX semantics the reading process will either see the full content before the rename op or the full content after the rename op. The writing process can replace the name (rename op) while the reading process reads the status file because its file descriptor still points to the old status file. Christian -- http://mail.python.org/mailman/listinfo/python-list
Re: How to safely maintain a status file
On 12/07/12 14:30:41, Laszlo Nagy wrote: You are contradicting yourself. Either the OS is providing a fully atomic rename or it doesn't. All POSIX compatible OS provide an atomic rename functionality that renames the file atomically or fails without loosing the target side. On POSIX OS it doesn't matter if the target exists. This is not a contradiction. Although the rename operation is atomic, the whole change status process is not. It is because there are two operations: #1 delete old status file and #2. rename the new status file. And because there are two operations, there is still a race condition. I see no contradiction here. On Posix systems, you can avoid the race condition. The trick is to skip step #1. The rename will implicitly delete the old file, and it will still be atomic. The whole process now consists of a single stop, so the whole process is now atomic. You don't need locks or any other fancy stuff. You just need to make sure that you flush the data and metadata correctly to the disk and force a re-write of the directory inode, too. It's a standard pattern on POSIX platforms and well documented in e.g. the maildir RFC. It is not entirely true. We are talking about two processes. One is reading a file, another one is writting it. They can run at the same time, so flushing disk cache forcedly won't help. On Posix systems, it will work, and be atomic, even if one process is reading the old status file while another process is writing the new one. The old file will be atomically removed from the directory by the rename operation; it will continue to exists on the hard drive, so that the reading process can continue reading it. The old file will be deleted when the reader closes it. Or, if the system crashed before the old file is closed, it will deleted when the system is restarted. On Windows, things are very different. Hope this helps, -- HansM -- http://mail.python.org/mailman/listinfo/python-list
Re: How to safely maintain a status file
Laszlo Nagy: This is not a contradiction. Although the rename operation is atomic, the whole change status process is not. It is because there are two operations: #1 delete old status file and #2. rename the new status file. And because there are two operations, there is still a race condition. I see no contradiction here. Christian Heimes li...@cheimes.de wrote: Sorry, but you are wrong. It's just one operation that boils down to point name to a different inode. For some reason you're assuming POSIX semantics, an assumption that Laszlo Nagy did not make. Ross Ridge -- l/ // Ross Ridge -- The Great HTMU [oo][oo] rri...@csclub.uwaterloo.ca -()-/()/ http://www.csclub.uwaterloo.ca/~rridge/ db // -- http://mail.python.org/mailman/listinfo/python-list
Re: How to safely maintain a status file
Sorry, but you are wrong. It's just one operation that boils down to point name to a different inode. After the rename op the file name either points to a different inode or still to the old name in case of an error. The OS guarantees that all processes either see the first or second state (in other words: atomic). POSIX has no operation that actually deletes a file. It just has an unlink() syscall that removes an associated name from an inode. As soon as an inode has no names and is not references by a file descriptor, the file content and inode is removed by the operating system. rename() is more like a link() followed by an unlink() wrapped in a system wide global lock. Then please help me understand this. Good case: process #1: unlink(old status file) process #1: rename(new status file) process#2: open(new status file) process#2: read(new status file) Bad case: process #1: unlink(old status file) process#2: open(???) -- there is no file on disk here, this system call returns with an error! process #1: rename(new status file) If it would be possible to rename + unlink in one step, then it would be okay. Can you please explain what am I missing? -- http://mail.python.org/mailman/listinfo/python-list
Re: How to safely maintain a status file
This is not a contradiction. Although the rename operation is atomic, the whole change status process is not. It is because there are two operations: #1 delete old status file and #2. rename the new status file. And because there are two operations, there is still a race condition. I see no contradiction here. On Posix systems, you can avoid the race condition. The trick is to skip step #1. The rename will implicitly delete the old file, and it will still be atomic. The whole process now consists of a single stop, so the whole process is now atomic. Well, I didn't know that this is going to work. At least it does not work on Windows 7 (which should be POSIX compatible?) f = open(test.txt,wb+) f.close() f2 = open(test2.txt,wb+) f2.close() import os os.rename(test2.txt,test.txt) Traceback (most recent call last): File stdin, line 1, in module WindowsError: [Error 183] File already exists I have also tried this on FreeBSD and it worked. Now, let's go back to the original question: This works well on Linux but Windows raises an error when status_file already exists. It SEEMS that the op wanted a solution for Windows -- http://mail.python.org/mailman/listinfo/python-list
Re: How to safely maintain a status file
Windows doesn't suppport atomic renames if the right side exists. I suggest that you implement two code paths: if os.name == posix: rename = os.rename else: def rename(a, b): try: os.rename(a, b) except OSError, e: if e.errno != 183: raise os.unlink(b) os.rename(a, b) Problem is if the process is stopped between unlink and rename there would no status file. Yes, and actually it does not need to be an abnormal termination. It is enough if the OS scheduler puts this process on hold for some time... But using a lock file, the problem can be solved. However in that case, reading a status file can be a blocking operation. -- http://mail.python.org/mailman/listinfo/python-list
Re: How to safely maintain a status file
Am 12.07.2012 19:43, schrieb Laszlo Nagy: Well, I didn't know that this is going to work. At least it does not work on Windows 7 (which should be POSIX compatible?) Nope, Windows's file system layer is not POSIX compatible. For example you can't remove or replace a file while it is opened by a process. Lot's of small things work slightly differently on Windows or not at all. Christian -- http://mail.python.org/mailman/listinfo/python-list
Re: How to safely maintain a status file
On Jul 12, 2:39 pm, Christian Heimes li...@cheimes.de wrote: Windows's file system layer is not POSIX compatible. For example you can't remove or replace a file while it is opened by a process. Sounds like a reasonable fail-safe to me. Not much unlike a car ignition that will not allow starting the engine if the transmission is in any *other* gear besides park or neutral, OR a governor (be it mechanical or electrical) that will not allow the engine RPMs to exceed a maximum safe limit, OR even, ABS systems which pulse the brakes to prevent overzealous operators from loosing road-to-tire traction when decelerating the vehicle. You could say: Hey, if someone is dumb enough to shoot themselves in the foot then let them... however, sometimes fail-safes not only save the dummy from a life of limps, they also prevent catastrophic collateral damage to rest of us. -- http://mail.python.org/mailman/listinfo/python-list
Re: How to safely maintain a status file
On Thu, 12 Jul 2012 15:05:26 +0200, Christian Heimes wrote: You need to flush the data to disk as well as the metadata of the file and its directory in order to survive a system crash. The close() syscall already makes sure that all data is flushed into the IO layer of the operating system. And some storage devices (e.g. hard drives, USB sticks) don't actually write data permanently even when you sync the device. They just write to a temporary cache, then report that they are done (liar liar pants on fire). Only when the cache is full, or at some random time at the device's choosing, do they actually write data to the physical media. The result of this is that even when the device tells you that the data is synched, it may not be. -- Steven -- http://mail.python.org/mailman/listinfo/python-list
Re: How to safely maintain a status file
On Fri, Jul 13, 2012 at 11:20 AM, Rick Johnson rantingrickjohn...@gmail.com wrote: On Jul 12, 2:39 pm, Christian Heimes li...@cheimes.de wrote: Windows's file system layer is not POSIX compatible. For example you can't remove or replace a file while it is opened by a process. Sounds like a reasonable fail-safe to me. POSIX says that files and file names are independent. I can open a file based on its name, delete the file based on its name, and still have the open file there. When it's closed, it'll be wiped from the disk. ChrisA -- http://mail.python.org/mailman/listinfo/python-list
Re: How to safely maintain a status file
On Fri, 13 Jul 2012 12:12:01 +1000, Chris Angelico wrote: On Fri, Jul 13, 2012 at 11:20 AM, Rick Johnson rantingrickjohn...@gmail.com wrote: On Jul 12, 2:39 pm, Christian Heimes li...@cheimes.de wrote: Windows's file system layer is not POSIX compatible. For example you can't remove or replace a file while it is opened by a process. Sounds like a reasonable fail-safe to me. Rick has obviously never tried to open a file for reading when somebody else has it opened, also for reading, and discovered that despite Windows being allegedly a multi-user operating system, you can't actually have multiple users read the same files at the same time. (At least not unless the application takes steps to allow it.) Or tried to back-up files while some application has got them opened. Or open a file while an anti-virus scanner is oh-so-slowly scanning it. Opening files for exclusive read *by default* is a pointless and silly limitation. It's also unsafe: if a process opens a file for exclusive read, and then dies, *no other process* can close that file. At least on POSIX systems, not even root can override a mandatory exclusive lock (it would be pretty pointless if it could), so a rogue or buggy program could wreck havoc with mandatory exclusive file locks. That's why Linux, by default, treats exclusive file locks as advisory (cooperative), not mandatory. In general, file locking is harder than it sounds, with many traps for the unwary, and of course the semantics are dependent on both the operating system and the file system. https://en.wikipedia.org/wiki/File_locking POSIX says that files and file names are independent. I can open a file based on its name, delete the file based on its name, and still have the open file there. When it's closed, it'll be wiped from the disk. One neat trick is to open a file, then delete it from disk while it is still open. So long as your process is still running, you can write to this ghost file, as normal, but no other process can (easily) see it. And when your process ends, the file contents is automatically deleted. This is remarkably similar to what Python does with namespaces and dicts: # create a fake file system ns = {'a': [], 'b': [], 'c': []} # open a file myfile = ns['a'] # write to it myfile.append('some data') # delete it from the file system del ns['a'] # but I can still read and write to it myfile.append('more data') print(myfile[0]) # but anyone else will get an error if they try another_file = ns['a'] -- Steven -- http://mail.python.org/mailman/listinfo/python-list
Re: How to safely maintain a status file
On Thursday 12 July 2012 23:21:16 Steven D'Aprano did opine: On Fri, 13 Jul 2012 12:12:01 +1000, Chris Angelico wrote: On Fri, Jul 13, 2012 at 11:20 AM, Rick Johnson rantingrickjohn...@gmail.com wrote: On Jul 12, 2:39 pm, Christian Heimes li...@cheimes.de wrote: Windows's file system layer is not POSIX compatible. For example you can't remove or replace a file while it is opened by a process. Sounds like a reasonable fail-safe to me. Rick has obviously never tried to open a file for reading when somebody else has it opened, also for reading, and discovered that despite Windows being allegedly a multi-user operating system, you can't actually have multiple users read the same files at the same time. Chuckle. That was one of the 'features' that os9 on the trs-80 color computer had back in the 80's, and it was clean and well done because of the locking model the random block file manager had in OS9 for 6809 cpu's, no relation to the Mac OS9 other than a similar name. That color computer has a separate, text only video card I could plug in and display on an 80 column amber screen monitor. When I wanted to impress the visiting frogs, I often did something I have never been able to do on any other operating system since, start assembling a long assembly language file on one of the screens on the color monitor, hit the clear key to advance to the amber screen and start a listing on it of the assemblers output listing file. Because the file locking was applied only to the sector (256 bytes on that machine) being written at the instant, the listing would fly by till it caught up with the assemblers output, running into the lock and then dutifully following along, one sector behind the assemblers output, until the assembly was finished. That was in 1986 folks, and in the year of our Lord 2012, 26 years later, I still cannot do that in linux. When I ask why not, the replies seem to think I'm from outer space. Its apparently a concept that is not even attempted to be understood by the linux code carvers. Something is drastically wrong with that picture IMO. (At least not unless the application takes steps to allow it.) Or tried to back-up files while some application has got them opened. That in fact, ran me out of the amiga business in 1999, a 30Gb drive failed on my full blown 040 + 64 megs of dram A2000. When the warranty drive arrived is when I found that due to file locks on the startup files, all of them involved with the booting of that machine, my high priced Diavolo Pro backup tapes didn't contain a single one of those files. The linux box with Red Hat 5.0 on it that I had built in late 1998 to see what linux was all about found space under that desk yet that evening and I never looked back. Or open a file while an anti-virus scanner is oh-so-slowly scanning it. Opening files for exclusive read *by default* is a pointless and silly limitation. It's also unsafe: if a process opens a file for exclusive read, and then dies, *no other process* can close that file. At least on POSIX systems, not even root can override a mandatory exclusive lock (it would be pretty pointless if it could), so a rogue or buggy program could wreck havoc with mandatory exclusive file locks. That's why Linux, by default, treats exclusive file locks as advisory (cooperative), not mandatory. In general, file locking is harder than it sounds, with many traps for the unwary, and of course the semantics are dependent on both the operating system and the file system. https://en.wikipedia.org/wiki/File_locking POSIX says that files and file names are independent. I can open a file based on its name, delete the file based on its name, and still have the open file there. When it's closed, it'll be wiped from the disk. One neat trick is to open a file, then delete it from disk while it is still open. So long as your process is still running, you can write to this ghost file, as normal, but no other process can (easily) see it. And when your process ends, the file contents is automatically deleted. This is remarkably similar to what Python does with namespaces and dicts: # create a fake file system ns = {'a': [], 'b': [], 'c': []} # open a file myfile = ns['a'] # write to it myfile.append('some data') # delete it from the file system del ns['a'] # but I can still read and write to it myfile.append('more data') print(myfile[0]) # but anyone else will get an error if they try another_file = ns['a'] Cheers, Gene -- There are four boxes to be used in defense of liberty: soap, ballot, jury, and ammo. Please use in that order. -Ed Howdershelt (Author) My web page: http://coyoteden.dyndns-free.com:85/gene is up! You just wait, I'll sin till I blow up! -- Dylan Thomas -- http://mail.python.org/mailman/listinfo/python-list
Re: How to safely maintain a status file
On Thu, 12 Jul 2012 23:49:02 -0400, Gene Heskett wrote: When I wanted to impress the visiting frogs, I often did something I have never been able to do on any other operating system since, start assembling a long assembly language file on one of the screens on the color monitor, hit the clear key to advance to the amber screen and start a listing on it of the assemblers output listing file. Because the file locking was applied only to the sector (256 bytes on that machine) being written at the instant, the listing would fly by till it caught up with the assemblers output, running into the lock and then dutifully following along, one sector behind the assemblers output, until the assembly was finished. That was in 1986 folks, and in the year of our Lord 2012, 26 years later, I still cannot do that in linux. Um, what you are describing sounds functionally equivalent to what tail -f does. When I ask why not, the replies seem to think I'm from outer space. Its apparently a concept that is not even attempted to be understood by the linux code carvers. You could certainly create a pair of cooperative programs, one which keeps a lock on only the last block of the file, and a tail-like reader which honours that lock. But why bother? Just have the assembler append to the file, and let people use any reader they like, such as tail. Or have I misunderstood you? -- Steven -- http://mail.python.org/mailman/listinfo/python-list
Re: How to safely maintain a status file
On Thursday, July 12, 2012 10:13:47 PM UTC-5, Steven D#39;Aprano wrote: Rick has obviously never tried to open a file for reading when somebody else has it opened, also for reading, and discovered that despite Windows being allegedly a multi-user operating system, you can#39;t actually have multiple users read the same files at the same time. You misread my response. My comment was direct result of Christian stating: (paraphrase) On some systems you are not permitted to delete a file whilst the file is open ...which seems to be consistent to me. Why would *anybody* want to delete a file whilst the file is open? Bringing back the car analogy again: Would you consider jumping from a moving vehicle a consistent interaction with the interface of a vehicle? Of course not. The interface for a vehicle is simple and consistent: 1. You enter the vehicle at location A 2. The vehicle transports you to location B 3. You exit the vehicle At no time during the trip would anyone expect you to leap from the vehicle. But when you delete open files, you are essentially leaping from the moving vehicle! This behavior goes against all expectations of consistency in an API -- and against all sanity when riding in a vehicle! Opening files for exclusive read *by default* is a pointless and silly limitation. It#39;s also unsafe: if a process opens a file for exclusive read, and then dies, *no other process* can close that file. Oh come on. Are you actually going to use errors or unintended consequences, or even Acts of God to defend your argument? Okay. Okay. I suppose IF the car spontaneously combusted THEN the passengers would be wise to jump out, leaving the vehicle to the whims of inertia. One neat trick is to open a file, then delete it from disk while it is still open. So long as your process is still running, you can write to this ghost file, as normal, but no other process can (easily) see it. And when your process ends, the file contents is automatically deleted. Well neat tricks aside, I am of the firm belief that deleting files should never be possible whilst they are open. * Opening files requires that data exist on disk * Reading and writing files requires an open file obj * Closing files requires an open file object * And deleting files requires that the file NOT be open Would you also entertain the idea of reading or writing files that do not exist? (not including pseudo file objs like StringIO of course!). Summary: Neat tricks and Easter eggs are real hoot, but consistency in APIs is the key. -- http://mail.python.org/mailman/listinfo/python-list
Re: How to safely maintain a status file
and then on startup read from tmp_file if status_file does not exist. But this seems awkward. It also violates your requirement -- since the crash could take place with a partial temp file. Can you explain why? My thinking was if crash took place when writing the temp file this would not matter because the status file would still exist and be read from. The temp file would only be renamed when fully written. -- http://mail.python.org/mailman/listinfo/python-list
Re: How to safely maintain a status file
Windows doesn't suppport atomic renames if the right side exists. I suggest that you implement two code paths: if os.name == posix: rename = os.rename else: def rename(a, b): try: os.rename(a, b) except OSError, e: if e.errno != 183: raise os.unlink(b) os.rename(a, b) Problem is if the process is stopped between unlink and rename there would no status file. -- http://mail.python.org/mailman/listinfo/python-list
Re: How to safely maintain a status file
Am 09.07.2012 07:50, schrieb Plumo: Windows doesn't suppport atomic renames if the right side exists. I suggest that you implement two code paths: Problem is if the process is stopped between unlink and rename there would no status file. Yeah, you have to suffer all of Windows' design flaws. You could add a backup status file or use a completely different approach. -- http://mail.python.org/mailman/listinfo/python-list
Re: How to safely maintain a status file
On Sun, 08 Jul 2012 22:57:56 +0200, Laszlo Nagy wrote: Yes, this is much better. Almost perfect. Don't forget to consult your system documentation, and check if the rename operation is atomic or not. (Most probably it will only be atomic if the original and the renamed file are on the same physical partition and/or mount point). On Unix, rename() is always atomic, and requires that source and destination are on the same partition (if you want to move a file across partitions, you have to copy it then delete the original). But even if the rename operation is atomic, there is still a race condition. Your program can be terminated after the original status file has been deleted, and before the temp file was renamed. In this case, you will be missing the status file (although your program already did something just it could not write out the new status). In the event of abnormal termination, losing some data is to be expected. The idea is to only lose the most recent data while keeping the old copy, rather than losing everything. Writing to a temp file then rename()ing achieves that. -- http://mail.python.org/mailman/listinfo/python-list
Re: How to safely maintain a status file
Richard Baron Penman richar...@gmail.com wrote: Is there a better way? Or do I need to use a database? Using a database would seem to meet a lot of your needs. Don't forget that Python comes with a sqlite database engine included, so it shouldn't take you more than a few lines of code to open the database once and then write out your status every few seconds. import sqlite3 con = sqlite3.connect('status.db') ... with con: cur = con.cursor() cur.execute('UPDATE ...', ...) and similar code to restore the status or create required tables on startup. -- Duncan Booth http://kupuguy.blogspot.com -- http://mail.python.org/mailman/listinfo/python-list
Re: How to safely maintain a status file
On 7/8/2012 2:52 PM, Christian Heimes wrote: You are contradicting yourself. Either the OS is providing a fully atomic rename or it doesn't. All POSIX compatible OS provide an atomic rename functionality that renames the file atomically or fails without loosing the target side. On POSIX OS it doesn't matter if the target exists. Rename on some file system types (particularly NFS) may not be atomic. You don't need locks or any other fancy stuff. You just need to make sure that you flush the data and metadata correctly to the disk and force a re-write of the directory inode, too. It's a standard pattern on POSIX platforms and well documented in e.g. the maildir RFC. You can use the same pattern on Windows but it doesn't work as good. That's because you're using the wrong approach. See how to use ReplaceFile under Win32: http://msdn.microsoft.com/en-us/library/aa365512%28VS.85%29.aspx Renaming files is the wrong way to synchronize a crawler. Use a database that has ACID properties, such as SQLite. Far fewer I/O operations are required for small updates. It's not the 1980s any more. I use a MySQL database to synchronize multiple processes which crawl web sites. The tables of past activity are InnoDB tables, which support transactions. The table of what's going on right now is a MEMORY table. If the database crashes, the past activity is recovered cleanly, the MEMORY table comes back empty, and all the crawler processes lose their database connections, abort, and are restarted. This allows multiple servers to coordinate through one database. John Nagle -- http://mail.python.org/mailman/listinfo/python-list
Re: How to safely maintain a status file
Please consider batching this data and doing larger writes. Thrashing the hard drive is not a good plan for performance or hardware longevity. For example, crawl an entire FQDN and then write out the results in one operation. If your job fails in the middle and you have to start that FQDN over, no big deal. If that's too big of a chunk for your purposes, perhaps break each FQDN up into top-level directories and crawl each of those in one operation before writing to disk. There are existing solutions for managing job queues, so you can choose what you like. If you're unfamiliar, maybe start by looking at celery. Michael On Mon, Jul 9, 2012 at 1:52 AM, Plumo richar...@gmail.com wrote: What are you keeping in this status file that needs to be saved several times per second? Depending on what type of state you're storing and how persistent it needs to be, there may be a better way to store it. Michael This is for a threaded web crawler. I want to cache what URL's are currently in the queue so if terminated the crawler can continue next time from the same point. -- http://mail.python.org/mailman/listinfo/python-list -- http://mail.python.org/mailman/listinfo/python-list
Re: How to safely maintain a status file
On Mon, Jul 9, 2012 at 8:24 PM, John Nagle na...@animats.com wrote: On 7/8/2012 2:52 PM, Christian Heimes wrote: You are contradicting yourself. Either the OS is providing a fully atomic rename or it doesn't. All POSIX compatible OS provide an atomic rename functionality that renames the file atomically or fails without loosing the target side. On POSIX OS it doesn't matter if the target exists. Rename on some file system types (particularly NFS) may not be atomic. Actually, ISTR that rename() is one of the few things on NFS that is atomic. http://bugs.python.org/issue8828 -- http://mail.python.org/mailman/listinfo/python-list
Re: How to safely maintain a status file
Am 09.07.2012 22:24, schrieb John Nagle: Rename on some file system types (particularly NFS) may not be atomic. The actual operation is always atomic but the NFS server may not notify you about success or failure atomically. See http://linux.die.net/man/2/rename, section BUGS. That's because you're using the wrong approach. See how to use ReplaceFile under Win32: http://msdn.microsoft.com/en-us/library/aa365512%28VS.85%29.aspx The page doesn't say that ReplaceFile is an atomic op. Christian -- http://mail.python.org/mailman/listinfo/python-list
Re: How to safely maintain a status file
On Jul 10, 6:24 am, John Nagle na...@animats.com wrote: That's because you're using the wrong approach. See how to use ReplaceFile under Win32: http://msdn.microsoft.com/en-us/library/aa365512%28VS.85%29.aspx I'm not convinced ReplaceFile is atomic: The ReplaceFile function combines several steps within a single function. An application can call ReplaceFile instead of calling separate functions to save the data to a new file, rename the original file using a temporary name, rename the new file to have the same name as the original file, and delete the original file. About the best you can get in Windows, I think, is MoveFileTransacted, but you need to be running Vista or later: http://msdn.microsoft.com/en-us/library/windows/desktop/aa365241(v=vs.85).aspx I agree with your suggestion of using something transactional that isn't bound to later Window versions, though. -- http://mail.python.org/mailman/listinfo/python-list
Re: How to safely maintain a status file
Am 08.07.2012 13:29, schrieb Richard Baron Penman: My initial solution was a thread that writes status to a tmp file first and then renames: open(tmp_file, 'w').write(status) os.rename(tmp_file, status_file) You algorithm may not write and flush all data to disk. You need to do additional work. You must also store the tmpfile on the same partition (better: same directory) as the status file with open(tmp_file, w) as f: f.write(status) # flush buffer and write data/metadata to disk f.flush() os.fsync(f.fileno()) # now rename the file os.rename(tmp_file, status_file) # finally flush metadata of directory to disk dirfd = os.open(os.path.dirname(status_file), os.O_RDONLY) try: os.fsync(dirfd) finally: os.close(dirfd) This works well on Linux but Windows raises an error when status_file already exists. http://docs.python.org/library/os.html#os.rename Windows doesn't suppport atomic renames if the right side exists. I suggest that you implement two code paths: if os.name == posix: rename = os.rename else: def rename(a, b): try: os.rename(a, b) except OSError, e: if e.errno != 183: raise os.unlink(b) os.rename(a, b) Christian -- http://mail.python.org/mailman/listinfo/python-list
Re: How to safely maintain a status file
What are you keeping in this status file that needs to be saved several times per second? Depending on what type of state you're storing and how persistent it needs to be, there may be a better way to store it. Michael On Sun, Jul 8, 2012 at 7:53 AM, Christian Heimes li...@cheimes.de wrote: Am 08.07.2012 13:29, schrieb Richard Baron Penman: My initial solution was a thread that writes status to a tmp file first and then renames: open(tmp_file, 'w').write(status) os.rename(tmp_file, status_file) You algorithm may not write and flush all data to disk. You need to do additional work. You must also store the tmpfile on the same partition (better: same directory) as the status file with open(tmp_file, w) as f: f.write(status) # flush buffer and write data/metadata to disk f.flush() os.fsync(f.fileno()) # now rename the file os.rename(tmp_file, status_file) # finally flush metadata of directory to disk dirfd = os.open(os.path.dirname(status_file), os.O_RDONLY) try: os.fsync(dirfd) finally: os.close(dirfd) This works well on Linux but Windows raises an error when status_file already exists. http://docs.python.org/library/os.html#os.rename Windows doesn't suppport atomic renames if the right side exists. I suggest that you implement two code paths: if os.name == posix: rename = os.rename else: def rename(a, b): try: os.rename(a, b) except OSError, e: if e.errno != 183: raise os.unlink(b) os.rename(a, b) Christian -- http://mail.python.org/mailman/listinfo/python-list -- http://mail.python.org/mailman/listinfo/python-list
Re: How to safely maintain a status file
On Sun, 8 Jul 2012 21:29:41 +1000, Richard Baron Penman richar...@gmail.com declaimed the following in gmane.comp.python.general: and then on startup read from tmp_file if status_file does not exist. But this seems awkward. It also violates your requirement -- since the crash could take place with a partial temp file. I'd suggest that, rather than deleting the old status file, you rename IT -- and only delete it IF you successfully rename the temp file. Yes, this is much better. Almost perfect. Don't forget to consult your system documentation, and check if the rename operation is atomic or not. (Most probably it will only be atomic if the original and the renamed file are on the same physical partition and/or mount point). But even if the rename operation is atomic, there is still a race condition. Your program can be terminated after the original status file has been deleted, and before the temp file was renamed. In this case, you will be missing the status file (although your program already did something just it could not write out the new status). Here is an algorithm that can always write and read a status (but it might not be the latest one). You can keep the last two status files. Writer: *create temp file, write new status info * create lock file if needed * flock it * try: *delete older status file * rename temp file to new status file * finally: unlock the lock file Reader: * flock the lock file * try: *select the newer status file * read status info * finally: unlock the lock file It is guaranteed that you will always have a status to read, and in most cases this will be the last one (because the writer only locks for a short time). However, it is still questionable, because your writer may be waiting for the reader to unlock, so the new status info may not be written immediatelly. It would really help if you could tell us what are you trying to do that needs status. Best, Laszlo -- http://mail.python.org/mailman/listinfo/python-list
Re: How to safely maintain a status file
Am 08.07.2012 22:57, schrieb Laszlo Nagy: But even if the rename operation is atomic, there is still a race condition. Your program can be terminated after the original status file has been deleted, and before the temp file was renamed. In this case, you will be missing the status file (although your program already did something just it could not write out the new status). You are contradicting yourself. Either the OS is providing a fully atomic rename or it doesn't. All POSIX compatible OS provide an atomic rename functionality that renames the file atomically or fails without loosing the target side. On POSIX OS it doesn't matter if the target exists. You don't need locks or any other fancy stuff. You just need to make sure that you flush the data and metadata correctly to the disk and force a re-write of the directory inode, too. It's a standard pattern on POSIX platforms and well documented in e.g. the maildir RFC. You can use the same pattern on Windows but it doesn't work as good and doesn't guaranteed file integrity for two reasons: 1) Windows's rename isn't atomic if the right side exists. 2) Windows locks file when a program opens a file. Other programs can't rename or overwrite the file. (You can get around the issue with some extra work, though.) Christian -- http://mail.python.org/mailman/listinfo/python-list
Re: How to safely maintain a status file
What are you keeping in this status file that needs to be saved several times per second? Depending on what type of state you're storing and how persistent it needs to be, there may be a better way to store it. Michael This is for a threaded web crawler. I want to cache what URL's are currently in the queue so if terminated the crawler can continue next time from the same point. -- http://mail.python.org/mailman/listinfo/python-list